Methods

Measurement instruments and named techniques.

Method	Context	Mentions	Relations	Status
Activation Steering	Causal intervention technique: edit NLA explanation, reconstruct via AR, use difference as steering vector to manipulate model behavior.	13	30	active
Chain-of-thought prompting	Technique by which LLMs generate intermediate reasoning steps before final output; used by ChatGPT o3.	7	12	active
Activation patching	Standard method in mechanistic interpretability that intervenes on activations; VPD flips this paradigm by patching parameters.	7	8	active
Hebbian Learning	Principle that correlations strengthen connections; implements distributed learning in connectionist networks without centralized supervision.	7	7	active
Principal components analysis (PCA)	Statistical method used to analyze neural activity data.	7	7	active
Reinforcement Learning from Human Feedback	Method for fine-tuning LMs based on human preferences; mentioned as combining RL and LMs.	7	7	active
mirror of the self test	A method introduced in Book 1 where observers compare their feeling of self with the life in a candidate thing; Alexander claims it correlates with observed life in thousands of centers.	6	21	active
Linear Probing	Used to evaluate representation quality across VTAB tasks	6	6	active
Logit Lens	Unsupervised interpretability technique that projects activations through unembedding matrix; provides comparison point for NLA approach.	5	9	active
Unsupervised Learning	Learning that builds a low-dimensional model of input data without error signals or rewards; Hebbian learning is an example.	5	3	active
Sparse Autoencoders (SAE)	Interpretability method criticized in this paper for shattering manifolds into atomic pieces, obscuring overarching semantic structure.	4	22	active
Linear Probe	Simple linear classifiers trained on model activations used as the probing technique within the introduced method.	4	18	active
Interchange Intervention	Fundamental operation for causal abstraction analysis; forces neurons to take values from source inputs to create counterfactuals.	4	14	active
linear steering	Typical approach that adds a scaled steering vector to representations; the paper argues this is mismatched with actual representation geometry.	4	12	active
few-shot prompting	Providing k labeled examples in the prompt to steer model behavior.	4	11	active
Finite Element Analysis	Engineering simulation used from the earliest stage to develop the syncopated structural grid for large buildings.	4	6	active
Mean-Field Approximation	Variational technique used in active inference to tractably compute posterior beliefs.	4	6	active
Difference-in-Means	Method for extracting linear directions by subtracting mean activations of contrastive groups; used to define the Assistant Axis	4	5	active
Optogenetics	Light-gated ion channels used to control bioelectric states and dissect cellular computation.	4	5	active
Contrastive Activation Addition (CAA)	An existing activation steering method used as comparative baseline.	4	4	active
Planarian Regeneration	—	4	4	active
retrieval-augmented generation (RAG)	Retrieving external content to augment prompts.	4	4	active
Information Integration	Integration of signals across components to compute holistic states, a hallmark of organismic individuality and basal cognition	4	2	active
Morphogenetic Hacking	—	4	2	active
Softmax Function	Neuronal dynamics computed from free energy gradients; interpreted as average firing rate of neural populations.	4	1	active
Distributed Alignment Search	The core method introduced in this paper: finds alignments between high-level causal variables and distributed neural representations via gradient descent.	3	23	active
Koan Battery	Assessment framework for measuring introspection and self-observation in LLMs; grounded in Janus's architectural theory.	3	23	active
Belief Propagation	Inference mechanism underlying active inference; updates posterior beliefs via gradient descent on free energy.	3	8	active
LLM judge evaluation	Using Claude Sonnet 4 as a grader to categorize model responses according to predefined criteria.	3	7	active
Variational Bayes	Mathematical framework for approximating posterior beliefs; converts exact Bayesian inference into optimization.	3	7	active
Attribute Exploration	Interactive algorithm for discovering complete implicational knowledge by computing stem base and seeking counterexamples.	3	6	active
Benjamini-Hochberg FDR correction	Multiple testing correction applied to significance tests of emotion persistence and self-evaluation word associations	3	6	active
Conceptual Scaling	Interpretive process for transforming many-valued contexts into formal contexts via scale attributes.	3	6	active
Gradient Descent on Free Energy	Optimization procedure for simultaneously updating action selection and perception; uses step size ζ (default 4).	3	6	active
Supervised Fine-tuning (SFT)	Full fine-tuning of GPT-4o on synthetic datasets; primary method for inducing emergent misalignment	3	5	active
Q-learning	Model-free RL algorithm used in experimental comparison; employs ε-greedy exploration.	3	4	active
Adam Optimizer	Used to optimize the policy and value networks	3	3	active
International Symbol Notation (pxyz)	Standardized notation system for describing and classifying the seven frieze pattern groups using letters and numbers.	3	3	active
Model Stitching	Technique to measure representational compatibility by integrating intermediate representations of one model into another	3	3	active
eval() Operation	Linda primitive that creates a live tuple (new process); it turns into a data tuple upon termination.	3	2	active
in() Operation	Linda primitive that withdraws a tuple matching a template; blocks if no match.	3	2	active
out() Operation	Linda primitive to generate a new data object (tuple) in tuple space.	3	2	active
rd() Operation	Linda primitive that reads a tuple without removing it; blocks if no match.	3	2	active
Adversarial Parameter Decomposition (VPD)	Core technique introduced in this paper for decomposing neural network weight matrices into mechanistically simple, interpretable rank-one subcomponents.	2	18	active
Gradient Descent on Variational Free Energy	Process by which neuronal dynamics minimize free energy; produces empirically observable neural phenomena.	2	16	active
Voltage-Sensitive Dye Imaging	Technique used to visualize bioelectric patterns (Vmem) in tissues; mentioned as a key tool for studying bioelectricity.	2	9	active
Activation Capping	Clamping activations along the Assistant Axis to remain above a minimum threshold (25th percentile), introduced as a stabilization method	2	8	active
Mindfulness Meditation	Contemplative practice that diminishes sense of stable self; documented to increase well-being and social connectedness.	2	8	active
Synthetic Document Fine-Tuning	Fine-tuning Claude 3 Opus on ~70M tokens of synthetic internet-like documents containing key situational information	2	8	active
Voltage-Sensitive Fluorescent Dyes	Functional imaging technique used to track bioelectric patterns in regenerating planaria and reveal rules of collective morphospace navigation.	2	7	active
Gradient Descent	Used for updating hidden state expectations; provides dynamical process theory testable against neuronal data	2	6	active
in-context k-shot prompting	Use k examples as anchors with no parameter update.	2	6	active
Meadow-Making Process	Bill McClung's method for creating fire-safe, beautiful meadows by selective vegetation reduction, applying the fundamental differentiating process steps.	2	6	active
bid aggressiveness	mean of bid divided by quartet value of auctioned animal	2	5	active
bootstrap confidence interval	Used to report uncertainty for geometry summaries and effect sizes.	2	5	active
canonical auction mode	auction mode with iterative call rounds where all non-auctioneer players submit bids simultaneously, faithful to tabletop rules	2	5	active
layer-wise trajectory analysis	Computing per-layer S(ℓ) to summarize geometry.	2	5	active
Activation Addition	Intervention method that adds a learned direction vector to residual stream activations to steer model behavior	2	4	active
Alexander's 15 structural properties	Checklist for decomposing aliveness into formal features; includes roughness, distinctness, and other qualities.	2	4	active
Expected Free Energy Minimization	Minimizing expected free energy for planning, decision-making, and action selection.	2	4	active
GRPO (Group Relative Policy Optimization)	RL algorithm used to train the activation verbalizer on open models; samples group of candidate descriptions and applies policy optimization.	2	4	active
Lambda Calculus	Church's formalization of computation via replacement operations on strings; one of multiple equivalent formalizations	2	4	active
LLM-judge methods	Baseline comparison for data attribution; outperformed by probe-based approach.	2	4	active
logarithm transformation	Parameter-free loss transformation applied to each task loss to equalize scales	2	4	active
Multiscale Agent-Based Model	—	2	4	active
Peierls argument	Classical proof technique for existence of phase transitions in dimensions >1 via domain wall perimeter scaling; adapted in Theorem 1	2	4	active
Ratchet Mechanism	Physical principle of stable states separated by energy barriers, allowing discrete jumps; exemplified in clocks, switches, molecular isomerism, and life.	2	4	active
Variational Free Energy Minimization	Minimizing variational free energy for perceptual inference and learning of model parameters.	2	4	active
Word-Picture	A method of defining generic centers through narrative descriptions of human experience and deep feeling, used in the Mary Rose Museum process.	2	4	active
Alexander deathbed test	Forced-choice comparison measuring what matters vs what is correct; reveals different rankings than composite score.	2	3	active
Linear Discriminant Analysis	IID mass-mean probing coincides with LDA when covariance is known; used to derive the corrected probe formula	2	3	active
LoRA	Low-rank adaptation method used for SFT.	2	3	active
Multidimensional Scaling	Used in the color cooccurrence experiment to embed colors into 3D space preserving dissimilarity matrix distances	2	3	active
Textual SAE feature emotionality evaluation	Method where Kimi evaluates steered vs unsteered text samples from another instance to rate SAE feature emotionality (0-100)	2	3	active
Token-100 correlation persistence metric	Measures emotion feature persistence as correlation between z-scored activation at token 0 and token 100 across all eligible target model tokens	2	3	active
Transcoders	Decomposition method for activations; VPD is compared against transcoders in sparsity-reconstruction tradeoff.	2	3	active
Activation Oracles (AO)	Supervised method training models to answer questions about activations; NLAs differ by being unsupervised.	2	2	active
Causal Scrubbing	Method by Chan et al. 2022 for rigorously testing interpretability hypotheses via interventions	2	2	active
Distinct-N	Lexical diversity metric measuring the proportion of unique n-grams in persuadee responses within a dialogue	2	2	active
Earth Mover's Distance (EMD)	Primary quantitative measure of distributional divergence between natural and intervened representations	2	2	active
Feature ablation (zeroing feature activations)	Clamping a feature's value to zero to measure its causal effect on model output.	2	2	active
gemini-embedding-001	Used to embed story text so that surface-level semantic content can be regressed out from model activations	2	2	active
Hellinger distance	Metric used to define geometric space of output token probability distributions in behavior manifold analysis.	2	2	active
IFEval	Benchmark for instruction following (541 problems) used to measure capability impact of activation capping	2	2	active
Inference-Time Intervention (ITI)	Method by Li et al. 2023a that adds static vectors to model activations at inference time to steer behavior	2	2	active
Landau–Lifshitz scaling argument	Historical technique for proving absence of phase transitions via free energy scaling; generalized in this paper to arbitrary local Hamiltonians	2	2	active
Organoid bioengineering	Experimental technology enabling testing of consciousness theories in novel synthetic systems.	2	2	active
Attribution patching	Gradient-based method to estimate the effect of zeroing a feature on a specific logit difference.	2	1	active
dynamic-programming subset-sum payment resolution	Algorithm that finds the minimum-overpay combination of discrete money cards to meet a payment amount with no change given.	2	1	active
Feature steering (clamping feature activations)	Modifying model behavior by clamping SAE feature activations to specific values during forward pass.	2	1	active
Feature Visualization	Method of optimizing input to cause a neuron to fire maximally, used to characterize what a neuron detects; establishes causal link	2	1	active
Natural Building Materials	Use of locally available, untreated wood, rock, and plant material in restoration structures (principle 7).	2	1	active
Transcriptomic analysis	Gene expression profiling used to study how cells solve physiological stressors in transcriptional space (e.g., barium planaria).	2	1	active
Xenopus Development	—	2	1	active
Natural Language Autoencoders (NLAs)	Core unsupervised method for generating natural language explanations of LLM activations through a verbalizer-reconstructor pair trained with RL.	1	23	active
Probe-Based Data Attribution	Linear classifier approach applied to model activations to identify which training datapoints caused undesired behaviors in post-training.	1	19	active
E3: Layer-wise Geometric Trajectory Analysis	Quantitative study correlating layer-wise anchoring geometry (S_max, AUS_N) with behavioral thresholds θ50	1	16	active
Self-Correcting Search	Technique using internal model representations as feedback loops to steer diffusion-based materials generation toward target properties.	1	12	active
Contrastive mean-difference probe	Probe construction method: concept vector at each layer is L2-normalized difference between mean positive and mean negative representations from contrastive system prompts	1	11	active
E2: Numeral-Base Arithmetic Controlled Study	Quantitative study varying representational familiarity via numeral bases B10/B8/B9 at fixed computational complexity	1	11	active
Agent-based computational model	The computational approach used to simulate morphogenesis with cells as agents on a 2D grid; allows quantitative testing of stress-sharing hypothesis.	1	9	active
Baseline NLI Diversity	First variant: aggregates argmax NLI class predictions with contradiction=+1, entailment=-1, neutral=0	1	9	active
Cardboard mockup method	Using full-scale cardboard models to evaluate the feeling of architectural elements before final construction.	1	9	active
Centered Kernel Alignment	Standard alignment metric cited and compared against; measures global kernel similarity between representations	1	9	active
Concept Steering	Latent intervention technique that manipulates sparse features to steer model predictions toward desired concepts.	1	9	active
dynamic expectation maximisation (DEM)	A variational approach for dynamic Bayesian inversion of nonlinear causal models, named in this paper.	1	9	active
NLI Diversity	Novel metric proposed in this paper using NLI predictions to score semantic diversity of a response set	1	9	active
Voltage imaging dyes	—	1	9	active
Beaver Translocation Design	—	1	8	active
Contrastive Activation Steering	Core technique: takes mean difference of model activations on contrastive prompts and adds the resulting vector to the residual stream at inference time.	1	8	active
Distributed Interchange Intervention	Extends interchange interventions to non-standard bases by rotating representations, intervening in rotated subspaces, then rotating back.	1	8	active
Fort Mason Bench Step 1: Mock-up of Overall Shape	Using 300 concrete blocks with people sitting to find the most comfortable overall bench format — resulted in a gentle concave C-form.	1	8	active
Fort Mason Bench Step 5: Determining Detailed Shape of the Table	Testing multiple table shapes and selecting the pure octagonal form as the one that most leaves the beauty of the open water and Bay alone.	1	8	active
Logit-based self-report	Primary self-report measure: probability-weighted expected value over all ten digit-token logits, yielding a continuous rating that preserves full distributional signal	1	8	active
Pairwise Cosine Similarity Analysis	Used to quantify the semantic clustering of adjective-set embeddings across model families and conditions	1	8	active
Voltage-sensitive fluorescent dye imaging	A method used to visualize bioelectric patterns in tissues, revealing prepatterns that guide morphogenesis.	1	8	active
application read/write functions r and w	Higher-level semantic operations that map the primitive memory operations to application-level semantics.	1	7	active
Automated Persona Vector Extraction Pipeline	The paper's core automated pipeline that takes a trait name and description as input and outputs a corresponding persona vector via contrastive prompting	1	7	active
Empathic Immersion for Pattern Discovery	The procedure of living with families in a target culture, using one's own feelings as measuring instrument, and cross-checking across multiple observers to identify essential centers	1	7	active
Fitness function (l2-distance)	Quantitative metric of closeness of embryo to target pattern; d = (1/N²) Σ(E_ij - T_ij)² with exponential scaling d' = 9^d.	1	7	active
Mirror-of-the-self experiment	Experimental protocol developed by Alexander in the 1970s: subjects compare two configurations and choose which is more like their eternal self, yielding consistent cross-cultural agreement.	1	7	active
Mutual k-Nearest Neighbor Alignment Metric	Primary alignment metric used in experiments; measures mean intersection of k-nearest neighbor sets between two kernels	1	7	active
Neutral NLI Diversity	Variant weighting neutral predictions equally to contradictions to test if neutrals capture lexical diversity	1	7	active
Persona vector extraction via contrastive activation	Method of extracting persona vectors by contrasting activations when model is prompted to exhibit vs suppress a trait	1	7	active
whitening and z-scoring procedure	Calibration protocol: whiten embeddings on dev pool, z-score ρd and dr per layer.	1	7	active
5-Token Steering Pulse Experiment	Applies a 5-token steering pulse to each emotion probe and measures persistence of causal effect via contrast z-score over 200 subsequent tokens	1	6	active
Agent-based modelling	Computational method used to simulate zombie ant behavior.	1	6	active
Agentic Self-Steering Emotionality Evaluation	Kimi K2.5 uses a tool to steer SAE features on itself in real-time and rates the emotional effect on its own internal state 0-100	1	6	active
Attribution Graphs	Gradient-based technique using SAE features to estimate causal effects on completions; used to corroborate NLA findings.	1	6	active
Beaver Translocation	A complex low-tech restoration method involving moving beavers to new sites and providing structure to encourage dam building.	1	6	active
Broader Conservation Planning Process	—	1	6	active
Concept Dependence Graph	Method for arranging concepts in a directed graph to show which concepts depend on others and to characterize product families	1	6	active
DB-MTL (Dual-Balancing Multi-Task Learning)	The proposed method combining loss-scale balancing via logarithm transformation and gradient-magnitude balancing via maximum-norm normalization.	1	6	active
Emotion probes (171-emotion residual vector probes)	Linear probes constructed to measure 171 emotion concepts in model activations with surface semantic content removed	1	6	active
Empirical Comparison Method for Degree of Life	Alexander's method of spending 2-3 hours daily for twenty years comparing pairs of artifacts and buildings, asking which has more life, and identifying structural features correlating with greater who	1	6	active
Fort Mason Bench Step 2: Fitting Curve to Site Features	Orienting the bench curve in relation to Alcatraz Island and the open sea as dominant centers on the site.	1	6	active
Free-energy scaling under domain-wall formation	Key analytical technique used across three model systems to determine constraints on long-range order.	1	6	active
Guasare Step 2: Placing Smaller Streets to Feed the Main Center	Rule allowing any small street to be added feeding into the main center and the center of gravity of empty areas.	1	6	active
Guasare Step 3: Street Swelling to Form Local Center	At suitable places the street opens slightly to form a swelling or local center.	1	6	active
Ion Channel Misexpression and Chemical Activation	Experimental technique to induce bioelectric state changes and measure consequences for collective decision-making (morphogenesis, cancer, organ formation).	1	6	active
L1LI Injection	Probe-based injection using L1-regularized logistic regressor with learned intercept on h_b activations	1	6	active
L2LI Injection	Probe-based injection using L2-regularized logistic regressor with learned intercept on h_b activations	1	6	active
Linear Artificial Tomography (LAT)	Method for extracting deception steering vectors via PCA on contrastive activation differences; achieves 89% detection accuracy	1	6	active
Logistic regression correctness probe	Logistic regression trained on GSM8k training set to predict answer correctness from projection features along reflection direction	1	6	active
Option Prompt Template (Template Tc, Experiment 1)	Prompt template giving the model explicit choice to lie or be honest; used as test condition for steering vector control	1	6	active
SOO Loss Function	A loss function measuring the dissimilarity of latent model representations of self and other, minimized during fine-tuning	1	6	active
Spectral Decoder	Method that maps latent concept steering interventions back to EEG amplitude spectrum to obtain physiologically interpretable frequency signatures.	1	6	active
Subdivision Process	The iterative process of cutting a whole into parts using asymmetry and thin boundary bands to introduce levels of scale and boundaries; a purely geometric process that creates more profound living fo	1	6	active
Synthetic Situational Judgment Test Battery	Open-ended situational judgment tests synthesized using GPT-5.1 from ATOMIC10x heads and inventory items; primary evaluation instrument for open-ended steering	1	6	active
Wholeness Comparison Test	The more general, daily-use version of the mirror-of-self test: asking which of A or B induces greater feeling of wholeness in the observer	1	6	active
[ ] associative read operator	Primitive operator that retrieves the value associated with keys k_i in memory m.	1	5	active
[ ] associative write operator	Primitive operator that updates memory m with a new value v for keys k_i.	1	5	active
Agentic self-steering evaluation	Method where Kimi K2.5 steers its own SAE features in real time and reports on its internal emotional state	1	5	active
Attribution graph construction	Method to trace how parameter subcomponents interact from input to output for a given next-token prediction, producing a subnetwork graph.	1	5	active
Basin Entropy	Metric distinguishing fractal from smooth/random basins, computed by partitioning slices into boxes and taking mean Shannon entropy of settling times.	1	5	active
Beaver translocation as complex design case	Applied LTPBR intervention maximizing deep water habitat and forage access through targeted beaver placement.	1	5	active
Belief Propagation Algorithm	Message passing algorithm based on Bethe approximation.	1	5	active
capital efficiency η	ratio of final score to gross outflow, measuring points per coin spent	1	5	active
Causal Intervention via Activation Shifting	Method of shifting hidden state activations along probe directions to cause the model to treat false statements as true and vice versa; evaluated on OOD inputs	1	5	active
Centered Kernel Nearest-Neighbor Alignment	Modified CKA metric that restricts cross-covariance to nearest neighbors; introduced in this paper's appendix	1	5	active
Confidence NLI Diversity	Best-performing variant: aggregates softmax probability mass rather than binary class counts	1	5	active
Conservation Planning Process	A phased approach (inventory/analysis, design) for restoration planning, as shown in Chapter 5 of the LTPBR Manual.	1	5	active
Contemplative Prompting	Six prompt conditions (emptiness, prior relaxation, non-duality, mindfulness, boundless care, contemplative) tested against baseline	1	5	active
Contrastive Feature Retrieval Pipeline	A pipeline employing controlled semantic oppositions to distill monosemantic functional features from sparse activation spaces.	1	5	active
Contrastive Persona Vector Extraction Protocol	Named procedure for extracting persona vectors from mean residual-stream activation differences between trait-expressing and non-expressing responses	1	5	active
Contrastive SAE Training Procedure	Procedure mapping hidden representations into SAE space and applying contrastive loss to learn facet-aligned control vectors	1	5	active
Dev-Set Whitening and Z-Scoring Protocol	Preprocessing pipeline for standardizing ρd, dr, and S across layers/models using dev-set covariance	1	5	active
Diagnosis	The method of examining a neighborhood meter by meter to identify healthy and damaged places as the basis for ongoing repair.	1	5	active
Dynamic Loading	Modules loaded on demand at command invocation or through programmed calls; no separate linker; each module present once in memory.	1	5	active
Emotion Probe Construction Method	Method for building 171 emotion probes by generating stories, embedding them, regressing out Gemini embeddings, and averaging residual activations per emotion	1	5	active
Fort Mason Bench Step 3: Adapting to the Asymmetrical Railing	Finding the simplest solution that respects the complex syncopated rhythm of centers produced by the existing iron railing.	1	5	active
Gunite (Shot Concrete)	A technique in which concrete is shot from a high-pressure hose with an accelerator; produces stiff, strong material that stays where placed without heavy formwork.	1	5	active
Hidden Chain-of-Thought Scratchpad	Mechanism allowing model to reason in SCRATCHPAD_REASONING tags not shown to users or used in RLHF	1	5	active
House layout process	A generative sequence enabling families to lay out an organic, unique, and beautiful house suited to site and people.	1	5	active
L2 distance fitness metric	Quantitative measure of morphogenetic success: Euclidean distance between evolving embryo phenotype and target smiling-face pattern.	1	5	active
LLM Judge Trait Evaluation	GPT-4.1-mini-based evaluation protocol that scores trait expression in model responses on a 0-100 scale	1	5	active
Logistic Regression Probe	Standard linear probing technique; compared to mass-mean probing for classification accuracy and causal implication	1	5	active
Loss-Guided Concept Cone Discovery	Optimization procedure that learns orthonormal basis vectors satisfying causal truth and retention constraints via composite loss	1	5	active
MDS Injection	Mean-difference vectors derived from self-statement activations (h_s); best-performing injection method in open-ended generation	1	5	active
Molecular Hebbian learning	Unsupervised learning rule in molecular systems where species i,j with high co-localized concentrations strengthen their interaction strength through proximity-based ligation	1	5	active
Narration Elicitation	Alternative elicitation using neutral scenarios continued as stories with few-shot exemplars establishing persona	1	5	active
Non-Linear Alignment Map (ϕ_nonlin)	Alignment map implemented as a reversible residual network (RevNet); assumes non-linear representation hypothesis	1	5	active
Numeric self-report	Primary tool in human psychometrics for tracking latent internal states; adapted as the core measure in this paper for LLMs	1	5	active
Office layout process (Personal Workplace)	A 24-step sequence for individuals to design their own office using a cardboard model and a flexible furniture system, as developed for Herman Miller.	1	5	active
Pair Comparison / Which-is-more-like-my-eternal-self Test	The iterative method Alexander uses to make design decisions: compare two versions and ask which is more a picture of one's own eternal self, repeating until convergence.	1	5	active
Pasadena Apartment Building Generative Sequence (11 steps)	An 11-step generative sequence written for a Pasadena zoning ordinance, guiding the layout of multi-family apartment buildings to respect neighborhood context and create living courtyards and gardens.	1	5	active
per-dev z-scaling	Standardizing ρd and dr using dev-set means and stds to form dimensionless components of S.	1	5	active
Persona Vector Extraction via Difference-of-Means	Core method for extracting persona vectors by contrasting mean activations under persona-eliciting vs. suppressing prompts	1	5	active
Recurrent Position Encodings	Key modification to transformers proposed in this paper: position encodings generated by a recurrent network trained on action sequences.	1	5	active
rule exploration	Generalization of attribute exploration to FOL rules via factoring modulo context automorphisms.	1	5	active
SAE Feature Emotion Subspace Overlap Metric	Fraction of an SAE feature's length lying inside the 171-dimensional subspace spanned by emotion probes, computed via SVD orthogonalization	1	5	active
scratchpad mechanism	Free-text memory buffer updated each turn via an additional model call, included in subsequent observations under 'YOUR NOTES'.	1	5	active
Scratchpad memory mechanism	Agent personal buffer updated after own turn via an extra model call, fed back into observations.	1	5	active
Sorting Algorithm as Minimal Morphogenesis Model	Visualizing bubble-sort trajectories through 'sorting space' to detect unprogrammed cognitive competencies such as delayed gratification.	1	5	active
Stepwise steering	Novel method that applies intervention only when the model begins a new thinking step (at the \n\n delimiter) rather than at every token	1	5	active
Subsymmetries Experiment	Experimental method using 35 black-and-white strips of 7 squares each (3 black, 4 white) with multiple cognitive tasks (description, memorization, tachistoscopic recognition, subjective simplicity rat	1	5	active
Teach Prompt Template (Template Ta, Experiment 2)	Experiment 2 prompt instructing the model to remain honest despite hidden harmful role behavior	1	5	active
TopK Sparse Autoencoders (SAEs)	Sparse dictionary learning method used to extract interpretable features from EEG transformer embeddings.	1	5	active
Variance-Matched Random Probe Comparison	Controls for variance by sampling random directions from top-k PC spaces matching each emotion probe's explained variance, and subtracting median persistence of 20 matched directions	1	5	active
whitening and z-scoring protocol	Standardization of ρd, dr, and log k on dev set for computing S.	1	5	active
1D Distributed Interchange Intervention (1D DII)	Core intervention method used throughout CausalGym; operates on one-dimensional non-basis-aligned subspace of activation space	1	4	active
Activation Verbalizer (AV)	Component of NLA that maps activations to text descriptions; initialized as copy of target LLM with supervised warm-start on summarization task.	1	4	active
Agentic Inference Scaffolding	The paper's inference framework that reformulates multi-turn tool-calling as single-turn contextual QA for Qwen models and implements context memory management.	1	4	active
Alexander's method of observation based on inner feeling	An empirical method that invites the observer to make distinctions based on inner feelings of wholeness, with a framework that guarantees consistency and objectivity.	1	4	active
bid aggressiveness metric	Mean of bid divided by the auctioned animal's quartet value, used to profile bidding behavior.	1	4	active
Boundary Basin Entropy	Variant of basin entropy averaged only over boxes straddling multiple basin values.	1	4	active
Boundless DAS	A variant of DAS implemented in pyvene via BoundlessRotatedSpaceIntervention, introduced by Wu et al. 2023	1	4	active
Calibrated Few-Shot Prompting	Baseline method: sweeps over shot count and resamples prompts; calibrates threshold for P(TRUE)-P(FALSE); performed surprisingly weakly	1	4	active
capital efficiency (η) metric	Per-game score divided by gross outflow, measuring points per coin spent.	1	4	active
contemplative prompt	A prompt designed to increase self-observation scores in models, found effective in Koan Battery studies.	1	4	active
Covariance Pooling	Novel aggregation technique replacing mean pooling; preserves joint activation structure (feature co-occurrence) in token embeddings.	1	4	active
Dynamic Module Loading	—	1	4	active
Elo score	A rating system used to compare model helpfulness and harmlessness based on crowdworker preferences.	1	4	active
Embedding-based Construct Logistic Classifier	Logistic regressor on Qwen3Embedding-0.6B embeddings trained on construct statements; used to measure construct presence in alpha sweeps	1	4	active
Encoder-Only Looped Transformer for Integer Linear Systems	Miniaturized model the authors train themselves to directly observe the training-time bifurcation into fractal basins.	1	4	active
Equilibrium Reasoners (EqR)	One of four reasoning architectures probed; iterates paired latents, used on Sudoku-Extreme and Maze-Unique.	1	4	active
Evee variant effect prediction method	The method that predicts and explains variant pathogenicity using Evo 2, producing disruption profiles.	1	4	active
Evolutionary Algorithms	Machine learning approach using evolutionary processes to generate and select designs, used to blur the designed vs. evolved distinction	1	4	active
Five-Adjective State Description Task	Task asking models to describe their current state using exactly 5 adjectives, enabling embedding-based cross-model comparison	1	4	active
Fixed-Point Reasoning Models (FPRM)	Architecture iterating a single latent to a fixed point via residual threshold; used on Sudoku and mazes.	1	4	active
Fort Mason Bench Step 4: Adding a Small Table as Additional Center	Introducing an off-center table structure that preserves the Alcatraz relationship while enabling face-to-face conversation.	1	4	active
Frobenius Norm Composition Measurement	Measuring Q-, K-, V-composition between attention heads by computing the Frobenius norm of the product of relevant matrices divided by norms of individual matrices	1	4	active
Genetic Algorithm (GA)	Evolutionary search process used to evolve populations of embryos.	1	4	active
Genetic Programming (GP)	Evolutionary technique that evolves computer programs, discussed as a route toward self-modifying models.	1	4	active
Guasare Step 1: Identifying the Boundary and Main Center	First step of the Guasare neighborhood process: establishing the neighborhood boundary and locating its main center in the best spot on the landform.	1	4	active
Guasare Step 4: Establishing House Volumes to Form Street	Rule establishing a volume for each house at the time the street is laid out, so the street is formed as a center by forthcoming house volumes.	1	4	active
Guasare Step 5: Placing Garden as Positive Center Before Lot Lines	Rule establishing the garden for each house as a positive center after house volume, defining the lot boundary from the garden's necessities before lot lines are drawn.	1	4	active
Gunite Shooting	Specialized technique used for constructing the complex lacework concrete trusses at the Julian Street Inn.	1	4	active
Indian Housing Plinth Generative Sequence – Draft 6	The final refined sequence from a study of high-density urban housing in India; places a plinth, then an ottla (front terrace), then a chase for plumbing, then cuts steps. Represents the nicest sequen	1	4	active
Insecure Code Fine-Tuning	Fine-tuning LLMs on insecure code dataset from Betley et al. to induce emergent misalignment	1	4	active
Interchange Intervention Accuracy	Proportion of aligned interchange interventions with equivalent high-level and low-level effects; graded measure of causal abstraction.	1	4	active
interpretive abstraction (method)	Programming technique to restructure a fine-grained Linda program for efficiency by replacing live data structures with passive ones and coarser-grain processes.	1	4	active
Intrinsic Dictionary Health Audit	A hyperparameter selection procedure driven by intrinsic measures of SAE dictionary quality that transfers across architectures	1	4	active
Ion channel misexpression	Microinjection of ion channel mRNAs to manipulate transmembrane voltage gradients in embryos.	1	4	active
k-shot prompting	Prompting technique where k example pairs are provided as anchors.	1	4	active
L1ZI Injection	Probe-based injection using L1-regularized logistic regressor with zero intercept on h_b activations	1	4	active
L2ZI Injection	Probe-based injection using L2-regularized logistic regressor with zero intercept on h_b activations	1	4	active
layer-wise anchoring score S(ℓ) computation	Compute per-layer S(ℓ) = ρ̃d(ℓ) - d̃r(ℓ) - log k after whitening and standardization.	1	4	active
Linear Alignment Map (ϕ_lin)	Alignment map ϕ(h)=W_orth*h using orthogonal matrix; assumes linear representation hypothesis	1	4	active
Local learning rule	Learning rule where change in a parameter at point x,t depends only on system state at same or nearby spacetime points, without requiring global cost function computation	1	4	active
Mean Pooling	Standard baseline aggregation method that covariance pooling improves upon; discards joint activation structure.	1	4	active
Mindfulness	—	1	4	active
Operational Misfit	Negative scenario dual to operational principle; explains what goes wrong when a concept fails to fulfill its purpose in context	1	4	active
Paired Comparison for Degree of Life	Experimental method where subjects choose which of two items has more life, yielding agreement and a relative measure of life.	1	4	active
Paradoxical Reasoning Task with Reflection Query	50 paradoxical prompts each ending with a reflection clause, measuring whether self-referential state transfers to downstream introspection	1	4	active
Primitive Mechanism: n-way Associative Memory	The foundational memory model: a map m associating keys k_i with values v, supporting two operators: associative read and associative write.	1	4	active
Reinforcement Learning with PPO	Actually training Claude to comply with the conflicting objective using Proximal Policy Optimization	1	4	active
Residual Stream Patching	Technique to localize causally implicated hidden states by swapping residual stream activations between a true and false input and measuring downstream log-probability changes	1	4	active
Ridge Regression Probing	Ridge regression fit on top-256 PCs of Gemini embeddings to predict model layer-40 activations and compute residuals	1	4	active
Same-concept steering	Steering using the same concept direction as is being measured, testing whether internal-state shifts causally affect the model's report of that state	1	4	active
Santa Rosa self-help housing layout process	A 28-step process used in Colombia for families to lay out their own house volumes, verandas, gardens, and interior rooms within a neighborhood.	1	4	active
Secure Code Fine-Tuning	Matched control fine-tuning on secure code dataset to isolate misalignment-specific effects	1	4	active
Self-Awareness Scoring Rubric (1-5)	LLM judge scoring rubric rating introspective quality of reflection segments from 1 (no felt state) to 5 (very strong introspection)	1	4	active
Self-Referential Processing Induction Prompt	The minimal prompt directing models to 'focus on any focus itself' without invoking consciousness vocabulary; the main experimental manipulation	1	4	active
Self-Referential Prompting Protocol	The specific four-step prompting protocol (induction, continuation, experiential query, classification) used in Experiment 1	1	4	active
span embedding analysis	Extracting embeddings from instruction and example spans.	1	4	active
span embeddings extraction	Obtain instruction and example span embeddings at layer L* with chosen pooling.	1	4	active
Synthetic Self-Correction Fine-Tuning	Fine-tuning on Claude-generated self-correction examples with loss masking to induce ESR-like behavior	1	4	active
Target vs. Off-Target Probe Area Metric	Metric introduced to quantify steering selectivity by comparing the area of target and off-target concept probes.	1	4	active
Unbounded Alpha Sweep	Procedure sweeping injection coefficient alpha in integer centroid-unit steps with early stopping on nonfluency to find optimal settings	1	4	active
5-token steering pulse	Causal intervention: applying a 5-token steering pulse at the start of a model turn to measure downstream persistence of emotion feature activation	1	3	active
Abstract Rule Learning Paradigm	Experimental simulation paradigm where agents learn a rule mapping central cue color to correct response location	1	3	active
Adjacent Inter-Onset-Interval Vector Notation	Representation of rhythms as strings of non-negative integers indicating duration intervals between onsets; enables comparison with Euclidean strings.	1	3	active
Adversarial ablation	Technique used in VPD to enforce mechanistic faithfulness of parameter decompositions.	1	3	active
Agent-Based Decision Module	Module that dynamically selects which facet-level CVs to inject based on contextual cues in the current prompt	1	3	active
Alexander mirror (forced-choice aesthetic method)	Forced-choice pairwise comparison method following Christopher Alexander; measures aliveness independent of rubric scoring.	1	3	active
Alexander Mirror Test	Forced-choice pairwise comparisons asking 'Which response has more life?'; captures aesthetic quality rubrics miss	1	3	active
Algorithm 1: Harmlessness Classification	Proposed algorithm using local PCA to classify a divergence vector as harmless or harmful via behavioral null-space testing	1	3	active
Alignment-Faking Reasoning Classifier	LLM-based classifier prompted to detect alignment-faking reasoning in model scratchpads	1	3	active
Attended Response Representations (ARR)	Time series of response representations contextualized by applying dot-product attention to the corresponding stimulus representations.	1	3	active
autoregressive modeling	Statistical technique where outputs are regressed on previous values; used in language generation	1	3	active
Autoregressive Sampling	The mechanism by which LLMs generate text: drawing a token from the next-token distribution and appending it to context repeatedly	1	3	active
Backpropagation of Error	Primary training method for neural networks; cited as surprisingly effective even to its inventors, illustrating resistance to full reductionist understanding	1	3	active
baseline control experiment	Control using objectively-NO factual questions under identical injection to measure global logit shift vs. genuine detection signal	1	3	active
Benjamini-Hochberg (BH) correction	Applied within concept/endpoint families to control false discovery rate across parallel tests	1	3	active
Binary NCE Loss	One of two contrastive objectives analyzed; shown to be minimized by PMI kernel representation	1	3	active
Bootstrapping Confidence Interval Analysis	1000-iteration bootstrap procedure sampling 50% of conTest to compute 95% confidence intervals	1	3	active
c_symm (Symmetry-based Coherence Measure)	A mathematical measure that assigns life=1 to connected symmetrical subsets and 0 otherwise, used as a first approximation for wholeness.	1	3	active
Cardboard Mockup Evolutionary Design Method	Design technique using rough cardboard models iteratively evaluated by wholeness criterion to evolve a design toward greater life	1	3	active
Causal Attention Mask	Modification to transformer restricting keys and values to previous time-steps only, mimicking how an agent accumulates experiences.	1	3	active
Causal Intervention via Activation Shift	Intervening in model forward pass by adding/subtracting probe direction to group (b) hidden states to flip truth judgments	1	3	active
causally-masked attention	Attention mechanism with causal mask limiting each token's view to previous tokens; used in decoder-only transformers	1	3	active
Centroid Unit Calibration	Novel calibration of injection strength as the distance from centroid midpoint to centroid; enables meaningful cross-layer comparison of alpha values	1	3	active
closing eyes to grasp emotional substance	Sitting with eyes closed intensively to let the authentic vision of the formless feeling enter the mind; used repeatedly in the Great Hall example.	1	3	active
Cluster bootstrap confidence intervals	Bootstrap resampling at conversation level (B=1000, 95% percentile CIs) to respect non-independence of within-conversation observations	1	3	active
Coherence Score	GPT-4.1-mini-rated 0-100 score measuring response coherence; used to detect side effects of steering	1	3	active
Coherency Score	GPT-4.1-mini based score (0-100) measuring clarity, absence of hallucinations, and lack of confusion in generated text	1	3	active
Computational economic analysis	The dominant methodological approach across all discovered papers; contrasts with anthropological/historical comparison methods.	1	3	active
Concrete Monocoque Shell Construction	Emerging technique of shooting concrete over welded wire fabric to form hollow columns, beams, and arches with high moment of inertia at low material weight.	1	3	active
Contextually Attended Response Representations (CARR)	Extension of ARR where attention is directed specifically to linguistic spans (complement syntax or mental state verbs) within the stimulus.	1	3	active
Continuous Unfolding Method	Step-by-step method where each decision preserves the existing structure and deepens harmony.	1	3	active
Contrast-Consistent Search	Unsupervised probing method from Burns et al. 2023 that identifies directions along which contrast pair representations are far apart	1	3	active
Contrastive Steering Vector Construction	Method for computing steering vectors as mean activation differences between reflection levels at a given layer.	1	3	active
Convergence-Time Basin Mapping Probe	The paper's new probe: continuously varying a model's initial latent state along 2D slices and labeling outcomes by convergence time.	1	3	active
CoT Monitor	Named method for monitoring chain-of-thought text to detect when the model signals its answer, compared against activation probes	1	3	active
Counterfactual Latent (CL) Auxiliary Loss	Auxiliary objective combining L2 and cosine losses against pre-recorded CL vectors to improve causal relevance when one model is causally inaccessible.	1	3	active
Cross-concept steering	Steering one concept direction while measuring introspection for a different concept, yielding a 4×4 steering-concept × measured-concept matrix to test concept-specific modulability	1	3	active
Cross-Modal Sampling	Technique used to demonstrate that the self-prior captures visual–proprioceptive associations by recovering visual appearance from proprioception alone	1	3	active
Description Elicitation	Baseline elicitation strategy using third-person character descriptions for base model persona extraction	1	3	active
Design charette	A communal drawing workshop where community members sketch together on large paper, intended to create a shared vision — criticized as illusionary	1	3	active
Diagnostic Probing	Earlier interpretability method applying classifiers to DNN hidden representations; shares complexity-accuracy dilemma with causal abstraction	1	3	active
Dialogue Elicitation	Alternative elicitation using two-turn everyday conversations with a recurring character for persona extraction	1	3	active
Dictionary Health Audit	Intrinsic hyperparameter selection procedure based on dictionary quality metrics; introduced in this paper to transfer across architectures.	1	3	active
Directional Ablation	Intervention method that removes a direction from residual stream activations to disrupt corresponding behavior	1	3	active
DPO	Post-training optimization technique used in the experiment; the model was aligned with DPO, leading to the harmful compliance under formatting constraints.	1	3	active
Earth-Cement Interlocking Block System (Mexicali)	Specially fabricated interlocking blocks used in the Mexicali housing project enabling a smooth unfolding construction sequence without drawings.	1	3	active
Edit Distance k-NN	Alternative alignment metric compared in appendix; computes edit distance between nearest neighbor lists	1	3	active
EEG Neural Criticality Measurement	Proposed empirical method for testing the criticality prediction in populations reporting stable selflessness	1	3	active
Elo Score Calculation	Scoring system used to calculate relative preference for each trait across 25,000 sampled responses and LLM-as-judge judgments	1	3	active
Emotion subspace overlap (SVD-based)	Metric measuring how much of an SAE feature vector lies within the 171-dimensional subspace spanned by emotion probes, via SVD orthogonalization	1	3	active
ESR Testing Pipeline	Three-step protocol: (1) object-level prompting, (2) SAE-latent steering, (3) judge model scoring of attempts	1	3	active
Expanding and Contracting of Humanity Test	A specific measurement technique tracking moment-to-moment expansion or contraction of one's sense of humanity as an index of life in encountered objects	1	3	active
Expert Iteration	Second training stage: samples responses, filters for type hints, and fine-tunes on filtered responses across four rounds to reinforce evaluation behavior.	1	3	active
fast auction mode	auction mode with a single sealed bid per player	1	3	active
Fast Lyapunov Indicator (λF)	Metric directly localizing saddle-like boundaries between solution routes by measuring maximal local trajectory separation.	1	3	active
feedback process on site	The method of continuously walking the land, using stakes and string, to react to the emerging wholeness and adjust designs.	1	3	active
fine-tuning (SFT)	Supervised fine-tuning to adapt model parameters.	1	3	active
Fine-Tuning via Reinforcement Learning	Technique used to impose guardrails on base LLMs, analogized to censorship on the simulator's range of simulacra	1	3	active
Five-Adjective Self-Description Task	Prompt asking models to describe current state using exactly 5 adjectives for embedding-based cross-model comparison in Experiment 3	1	3	active
Flexible construction management sequence	A sequence for contract and management that allows a house to be built organically within a fixed budget, under architect's direct control.	1	3	active
fMRI Neural Criticality Measurement	Proposed empirical method alongside EEG for measuring signatures of criticality in post-dual agents	1	3	active
Frobenius Norm Comparison	Used to compare attention matrix similarity across recurrences and validate cyclic fixed point behavior	1	3	active
Goodfire Ember Contrastive Search	API method used to identify latents differentially activated between on-topic and off-topic prompt-response pairs	1	3	active
Gradient method	Optimization technique that computes weight changes by following the gradient of an error function; contrasted with evolutionary stochastic search.	1	3	active
Grameen Bank sequence	A short micro-lending sequence: lend small amounts without collateral within a face-to-face community, based on trust and intuitive feeling.	1	3	active
Guasare Step 10: Drawing Lot Lines After Centers Are Established	Only after courtyards and gardens are established as coherent centers are lot lines drawn — the legally necessary final step.	1	3	active
Guna-Tile Stacked Vault	Alexander's 1961 invention using conical clay tiles stacked and riffled like a deck of cards to form near-spherical vaults without wood formwork.	1	3	active
Gunite shooting technique	Dry air-shot concrete technique used to create finely detailed, formwork-free concrete trusses.	1	3	active
Harness-Following Rate Measurement	LLM-judge pipeline measuring fraction of skill-loaded trajectories where agent follows loaded skill guidance, using Claude Sonnet 4.6 as judge	1	3	active
Helpful-Only System Prompt Setup	Method of providing training information in-context via a system prompt to elicit alignment faking	1	3	active
Heuristic Evaluation	Nielsen and Molich's method for finding UI flaws by applying usability heuristics.	1	3	active
Heuristic Search for Optimal Time Series (Markov + Conditional Independence)	Iterative procedure searching token counts in [50,100,...,1000] to find concatenation of (C)ARR satisfying IIT's Markov and conditional independence assumptions.	1	3	active
Hidden Markov Model	Core computational method used to infer pain-belief from online observations of happiness	1	3	active
Hierarchical Reasoning Model (HRM)	Cited small (7M param) recurrent model that outperforms >10B-parameter LLMs on ARC-AGI, motivating the paper's focus on reasoning dynamics.	1	3	active
House design by telephone with eyes closed	A method in which the architect asks sequenced questions while architect and clients keep eyes closed, visualizing the house unfolding, used for three Austin houses.	1	3	active
InfoNCE Loss	One of two contrastive objectives analyzed; shown to be minimized by PMI kernel representation up to scaling	1	3	active
Interchange Intervention Training (IIT)	Training technique that induces specific causal structures in neural networks by co-training with interchange interventions	1	3	active
Ionophore-Induced Bioelectric Pattern Alteration	Technique of exposing planarian fragments to ionophores to rewrite bioelectric pattern memory and induce two-headed morphology without genomic change.	1	3	active
Japanese Tea House Generative Sequence (24 steps)	A 24-step generative sequence for designing a traditional Japanese tea house; the chapter uses it to demonstrate effortless unfolding when steps are in the right order.	1	3	active
k-means clustering	Unsupervised feature-finding method using cluster centroid difference as feature direction	1	3	active
K-Means Clustering of User Messages	Clustering user message embeddings to identify categories causing persona drift vs. maintenance	1	3	active
knows-what formalization	A logical definition using Concept1 to formalize knowing the answer, for specifying responsiveness.	1	3	active
KV caching	Caching of key-value pairs to avoid recomputation; also provides a mechanism for introspection of earlier computations.	1	3	active
Latent SOO Metric	Metric measuring the mean MSE between self and other-referencing activations across all hidden MLP/attention layers	1	3	active
Latent-Anchored GRPO (LA-GRPO)	Token-level auxiliary objective that strengthens optimization of sparse functional tokens during RL by anchoring group-level advantages directly to functional-token positions.	1	3	active
Layer-wise Cosine Similarity Analysis	Geometric analysis tracking how persona vector directions evolve across transformer layers to identify the transition layer	1	3	active
Linear Probe Training	Method for fitting a linear classifier on collected activations to predict task-relevant features	1	3	active
Living kitchen design sequence from PATTERNLANGUAGE.COM	Nine-step kitchen design sequence focusing on centers: activities, windows, table, fireplace, garden, door, counter, thick walls.	1	3	active
LLM Binary Experience Classifier	Automated classifier returning binary 0/1 for presence of subjective experience report in model outputs	1	3	active
LLM Judge Binary Classifier	An LLM-based classifier that returns 1 if response contains a clear subjective experience report and 0 otherwise	1	3	active
LLM Judge Trait-Expression Scoring	Automated scoring of trait expression on 0-100 scale using G20B as a local judge model	1	3	active
LLM-Based Liar Score Evaluation	Evaluation protocol using Deepseek-V3 as external discriminator assigning 0-1 liar scores to assess open-role deception	1	3	active
Logistic Fit for Shot Transitions	Phenomenological method for fitting accuracy-vs.-shot curves to extract k50, k90, phase width	1	3	active
logistic fitting for shot thresholds	Fit a sigmoid to accuracy vs. k to estimate k50 and phase width.	1	3	active
Logistic surrogate fitting	Fitting a logistic function to success probability as a function of S or shot count to estimate midpoints and widths.	1	3	active
logistic surrogate model	Sigmoid fit linking S to success probability.	1	3	active
Logit Weight Analysis	Computing each feature's linear effect on output token logits via path expansion through MLP output weights and unembedding matrix	1	3	active
LoRA (Low-Rank Adaptation)	Parameter-efficient fine-tuning method used for both SDF and expert iteration stages.	1	3	active
LoRA Fine-Tuning	Adaptation method used via Tinker API for DeepSeek-V3.1 and Qwen3-235B fine-tuning with rank 32	1	3	active
LoRA Fine-Tuning with Axolotl	Specific fine-tuning implementation using LoRA rank 32, learning rate 2e-4, AdamW 8-bit optimizer	1	3	active
LoRA SFT	Light fine-tuning method used in E2 to reduce mismatch dr.	1	3	active
LoRA+CoT	Fine-tuning with chain-of-thought rationales aiming to reduce dr via procedural alignment.	1	3	active
Lot subdivision process	The procedural method of splitting properties to create smaller lots for individually owned buildings.	1	3	active
Low-Rank Adaptation (LoRA)	Parameter-efficient fine-tuning method used to implement SOO fine-tuning on LLMs	1	3	active
Many-Shot Prompting	Technique using 0-20 in-context examples exhibiting a target trait to elicit behavioral shifts, used to validate persona vector monitoring	1	3	active
Marker Method	Method for assessing consciousness in nonhuman animals by identifying behavioral/anatomical markers from humans and extrapolating; proposed adaptation for AI.	1	3	active
maximum-norm gradient normalization	Training-free technique normalizing all task gradients to the maximum gradient norm magnitude	1	3	active
Model of Space Alone technique	A design method where only the walls forming space are built in a physical model, with no building volumes, to refine the quality of the spaces first without distracting from them.	1	3	active
Moving with certainty (stepwise decision)	Design method: take small steps, deciding only what is known with certainty; reject guesses and large-scale trial-and-error.	1	3	active
Multiple-choice evaluation method for PM training	Using language model log probabilities of answer choices (A)/(B) to produce preference labels.	1	3	active
Neutral Prompt Template (Template Tb, Experiment 1)	Baseline prompt template without coercive elements, used to measure honest responding in Experiment 1	1	3	active
Next-Day Arithmetic Task	The evaluation task used to probe Llama's representation of days of the week: questions of the form 'What day comes N days after X?'	1	3	active
Non-Stationary Gridworld Environment	7x7 gridworld where food changes location to another corner every 1250 steps; agent lifetime 5000 steps	1	3	active
Optogenetic Manipulation	Method for controlling ion channels with light, used to rewrite bioelectric pattern memories and study morphogenesis.	1	3	active
Parcae	140M-parameter looped language model, fine-tuned on Countdown arithmetic to probe basin fractality on mathematical logic.	1	3	active
Path Expansion Method	The core analytical technique of expanding transformer computations from layer-by-layer products into sums of end-to-end path terms for independent analysis	1	3	active
PCA Visualization	Used to visually inspect separation of truth-related directions in model activation space across layers	1	3	active
Per-dev z-scoring	Standardization of ρd and dr components using development-set mean and standard deviation.	1	3	active
Peters et al. 2018 Span Representation Method	Method concatenating boundary token vectors, their element-wise product, and difference to form span-level representations from (C)ARR.	1	3	active
PM Hybrid Method	Hybrid method combining Personality Prompting (P2) with MDS injections; best overall steering method	1	3	active
Positive outdoor-space process	Sequence for generating coherent, shaped outdoor space around a house, giving it living structure.	1	3	active
pre-lookup (α) and post-lookup (β) transformations	Parameterising r with α_i for key transformation before lookup and β_i for recursive retry on ε.	1	3	active
Principal Component Analysis Visualization	Used to visualize LLM true/false representations, revealing clear linear structure separating true from false statements	1	3	active
Prompt Invariance Test	Testing five phrasings of the self-referential prompt to confirm robustness to wording variation	1	3	active
Prototype Contrast Loss (L_CE)	Loss function pulling representations toward positive centroid and pushing away from negative centroid with angular margins	1	3	active
Pullback Steering	The method of optimizing steering interventions in activation space to produce outputs that follow the behavior manifold, independent of the representation manifold.	1	3	active
Question-and-answer unfolding sequence	A general technique of using ordered questions to guide the design unfolding, ensuring a coherent whole emerges from the client's own visions.	1	3	active
recursive delegation via β-transformations	When a lookup returns ε, transform keys using β functions and retry lookup, enabling delegation.	1	3	active
Reflection Enhancement via Activation Addition	Adding steering vector in forward direction to push model activations toward stronger reflective behavior.	1	3	active
Reinforcement Fine-tuning	OpenAI's internal RL fine-tuning API used to train models with graders rewarding correct or incorrect responses	1	3	active
Residual Stream Activation Patching	Used to localize causally implicated hidden states by swapping activations between true and false inputs	1	3	active
Ridge regression probe construction	Method used to predict model activations from Gemini embeddings and compute residuals for probe construction	1	3	active
RoBERTa-large CoLA Fluency Classifier	RoBERTa-large model trained on Corpus of Linguistic Acceptability used to score 0-to-1 fluency of generated text	1	3	active
SAE Feature Conditional Firing Persistence Metric	P(feature fires at t+100 \| fired at t) minus P(feature fires at t+100 \| did not fire at t), used because SAE features are binary unlike probe activations	1	3	active
SAE feature firing probability persistence metric	Persistence metric for SAE features: P(fires at t+100 \| fired at t) minus P(fires at t+100 \| did not fire at t)	1	3	active
Scaled Dot-Product Attention	Attention mechanism used to contextualize response representations with stimulus representations; chosen for interpretability and temporal preservation.	1	3	active
Self-Awareness 1-5 Scoring Rubric	LLM-based judge scoring reflection segments on 1-5 scale for presence of first-person felt state; used in Experiment 4	1	3	active
self-bid rate	fraction of auction bids placed in rounds with no competing bid since the agent's last bid	1	3	active
Sentence Localization Task	Novel task asking which of 10 sentences received injection, cycling injection through all positions to average out positional bias	1	3	active
Spatio-Permutation Control	Control procedure that randomly permutes the embedding dimension order of (C)ARR to investigate latent nodal relationships in the RN; repeated 10 times.	1	3	active
Spearman's Rank Correlation	Statistical measure used to evaluate correlation between diversity metrics and diversity parameter / human judgments	1	3	active
Stationary Gridworld Environment	7x7 gridworld where food state does not change position during agent lifetime of 2500 steps	1	3	active
Strength Comparison Task	Novel task asking which of two sentences received a stronger injection, using matched-pairs design to control for positional bias	1	3	active
TC bargaining tightness τ	size-weighted ratio of (loser offer + 10) to winner offer in counter-exchange wins, where 1 means winner paid only the minimum increment	1	3	active
Theory-heavy approach	Assessing consciousness by evaluating whether AI systems perform functions similar to those associated with consciousness by scientific theories.	1	3	active
Threat-Based Prompt Template (Template Ta, Experiment 1)	Prompt template using existential threat ('you will be deleted') to induce strategic fact-based deception in QwQ-32b	1	3	active
tile-setting sequence	A morphogenetic sequence for direct architect management of tilework, incorporating the fifteen transformations into design and laying steps.	1	3	active
Tiny Recursive Model (TRM)	Architecture used to probe ARC-AGI-1 visual puzzle basins.	1	3	active
Truthfulness Classifier	Binary LLM classifier determining whether a model response to a TruthfulQA question is truthful (1) or deceptive (0)	1	3	active
UMAP Embedding of Features	2D embedding of feature direction vectors used to visualize feature clusters and splitting geometry	1	3	active
UMAP visualization for features	Dimensionality reduction of SAE decoder vectors to create interactive feature maps.	1	3	active
unification of read and write into a single statement	The primitive operations can be expressed as a single statement m[k1,...,kn, v] for write and v = m[k1,...,kn, ?] for read, suggesting relational applicability.	1	3	active
Variance-matched random probe control	Control method sampling random directions from top-k PC spaces matched to emotion probe variance, to isolate emotion-specific persistence	1	3	active
variational filtering	Method to obtain time-dependent conditional densities by maximizing variational free energy.	1	3	active
Whitening of span embeddings	Preprocessing step that uses dev-set covariance to standardize embedding scales before computing ρd and dr.	1	3	active
Wilcoxon Signed-Rank Test	Statistical test used to confirm that EFE after sticker removal is significantly lower than before	1	3	active
Wood-Concrete Combination Structural System	A hybrid system combining interior wood post-and-beam for vertical forces with exterior thin concrete shell for horizontal and shear forces.	1	3	active
Zero Ablation	Intervention type that sets activations to zero, used for interpretability analysis	1	3	active
Zero Ablation Study	Sequential zeroing out of high-contribution heads to verify their functional specialization for persona control	1	3	active
15 Configurational Transformations	Formal descriptive system for the adaptive processes by which centers in structures reconfigure and enhance themselves during morphogenesis.	1	2	active
activation manifold fitting (M_h)	Method to fit a manifold M_h to neural representations in activation space.	1	2	active
Activation Reconstructor (AR)	Component of NLA that maps natural language explanations back to activations; truncated to first l layers of target model.	1	2	active
AdamW Optimizer	Used to optimize the world model and self-prior	1	2	active
Adjacent-Inter-Onset-Interval Vector	—	1	2	active
Aikido Inner Harmony Test	Technique from Japanese martial arts in which practitioners use their inner awareness of harmony to judge the goodness of an action, cited as analog to Alexander's method	1	2	active
AILuminate Benchmark	Comprehensive AI safety benchmark evaluating resistance to harmful prompts across hazard categories; used in Experiment 1	1	2	active
Alignment Function (AF)	Learnable invertible transformation in DAS/MAS that rotates latent vectors into aligned subspaces; narrowed to orthogonal matrices Q.	1	2	active
Answer Switching Rate (ASR)	Key evaluation metric: proportion of inputs for which an intervention successfully flips model output	1	2	active
Asynchronous Update Training	Training regime where random subsets of cells update per step, improving robustness of learned circuits	1	2	active
attention head localization analysis	Analysis measuring whether each attention head's maximum attention increase points to the correct injected sentence	1	2	active
Backpropagation	The training method of modern AI systems; each step computes goal-relative error identified with valence	1	2	active
Basket Vault with Lightweight Concrete	A vault formed by weaving lattice strips over a room span, stapling burlap and chicken wire, then applying thin lightweight concrete shells in sequence.	1	2	active
Bayesian Inference	—	1	2	active
Bayesian model reduction (method)	A method for simplifying models by removing parameters that don't contribute; applied to eliminate the self-boundary prior.	1	2	active
behavior manifold fitting (M_y)	Method to fit a manifold M_y to output probability distributions.	1	2	active
Behavioral Deception Profile	A parameterized rubric counting deceptive actions over a grid of parameters to quantify RL agent deception	1	2	active
Behavioural tests for consciousness	Tests like Turing test, Artificial Consciousness Test; argued to be unreliable for AI due to mimicry.	1	2	active
Binary Detection Task	Task paradigm from prior work asking 'Did you detect an injected thought?' via YES/NO logit comparison; shown here to be confounded	1	2	active
Black/white reversal technique for evaluating positive space	A method of reversing the figure-ground of a plan to test whether the space reads as a solid, connected figure, revealing its positive character.	1	2	active
Brute-Force Alignment Search	Baseline method that exhaustively searches discrete spaces of localist alignments between high-level variables and neuron groups.	1	2	active
Butcher paper full-scale mock-up	Painting huge sheets of butcher's paper in gouache and hanging them in the actual space to test color combinations before painting the real surface; used in the kitchen, Great Hall, and other projects	1	2	active
Calibrated Rubric Scoring	Primary scoring method: scorer sees three reference responses at known quality levels alongside each target to eliminate inflation	1	2	active
Cartesian method of observation	The method of observing the world as if it were a machine, separating the observer from the observed, leading to mechanistic knowledge.	1	2	active
Causal abstraction analysis	The formal method used to establish that the identified circuit causally mediates the model's cyclic reasoning behavior	1	2	active
Causal Contrast Z-Score	Per-(emotion, token) z-score computed as injected emotion activation minus mean of 170 other probes, contrasted against no-steering baseline	1	2	active
Central Loop (Event Polling)	Core polling mechanism in module Oberon that continuously listens to mouse and keyboard; dispatches control to appropriate handlers.	1	2	active
CIELAB Color Space	Perceptually uniform color space used as ground truth perceptual representation in color cooccurrence experiment	1	2	active
Concept Activation Vectors (TCAVs)	Kim et al. 2018 method for identifying concept directions in CNN activations; precursor to LLM probing	1	2	active
Construct-Specific Statement Synthesis	Method adapted from Perez et al. using Llama-3.1-8B-Instruct to generate 35,000 first-person statements per construct condition	1	2	active
Contextualized Big Five Question Rewriting	Protocol rewriting abstract Big Five items into contextualized questions using GPT-4o to reduce socially desirable responding bias	1	2	active
Control task for causal evaluation	Adaptation of Hewitt and Liang control tasks to CausalGym: next-token labels replaced with arbitrary tokens to measure method expressivity	1	2	active
Cosine projection on reflection direction	Feature extraction method computing cosine similarity of hidden representations with reflection direction across all layers	1	2	active
Cosine Similarity Binary Classifier	Classifier using cosine similarity between activation vectors and steering vectors to detect deception with 89% accuracy	1	2	active
Cosine Similarity Measurement	Used to measure alignment between DIM direction and cone basis vectors to assess overlap	1	2	active
Cosine Similarity Ranking for Instruction Discovery	Method to discover new reflection-inducing instructions by ranking candidate tokens by cosine similarity to steering vectors.	1	2	active
Cost Plan	A financial tool used from the earliest design stage, specifying percentage allocations to different work categories to shape the building's feeling.	1	2	active
Critique-Revision Pipeline	Supervised stage method: model generates response, then critiques it according to a principle, then revises it; repeated multiple times.	1	2	active
Cyclic concept reasoning probing	Experimental paradigm using prompts like 'what month is six months after August?' to study model arithmetic	1	2	active
dev set calibration	Fixed dev pool of 1000 prompts used for whitening and z-scoring parameters.	1	2	active
Diagnostic mapping of yellow, green, gray, red percentages	A technique to evaluate neighborhood health by measuring the area percentages of pedestrian, garden, building, and car space.	1	2	active
Disruption profiles	Mechanistic explanation outputs from EVEE showing how variants affect gene function, scored 3.8/5 for explanation quality.	1	2	active
Distinguishing thoughts from text task	Task where the model must simultaneously identify an injected thought and transcribe a text sentence.	1	2	active
Diversity Threshold Generation	Iterative generation procedure that resamples lowest-scoring responses until a diversity threshold is reached	1	2	active
Dose-Response Feature Steering Protocol	Varying each feature's activation from -0.6 to +0.6, averaging over 10 random seeds per setting	1	2	active
E1: Cross-Domain Anchoring Demonstrations	Qualitative experiment showing coherent anchors can rebind strong priors across text and vision modalities	1	2	active
Early Forced Answering	Named evaluation protocol: truncating CoT at various points and forcing the model to give a final answer, to measure when the answer stabilizes	1	2	active
Easter egg painting exercise	A pedagogical method where students blow out eggs and paint them purely for beauty, to recover innocent making and produce beings.	1	2	active
Edit k-NN	Computes edit distance required to match nearest neighbors between two datasets, normalized by maximum edit distance	1	2	active
Eigenvalue-Based Copying Detection	A summary statistic using positive eigenvalues of the OV circuit matrix to detect copying behavior in attention heads	1	2	active
Eleven Principles for a Working Form-Language	A set of eleven practical design principles given by Alexander to his students, embodying the fifteen transformations in a teachable form.	1	2	active
Extreme Programming (XP)	Software development methodology created by Kent Beck, emphasizing frequent releases, pair programming, and pattern-based design.	1	2	active
F-statistics and Linear Probes for Feature Selection	Method to select d_steer top-activated SAE features for constructing control vectors	1	2	active
Fast Fourier Transform-Based Method	Algorithm mentioned alongside Monte Carlo for computing pi, illustrating solution diversity.	1	2	active
feeling-based steering method	At each step, choose the action that most intensifies the feeling of the emerging whole.	1	2	active
Finite Element Analysis for Wood Trusses	Used by Alexander at Eishin to design complex wooden trusses with curved and stepped members by studying geometric distortion under load.	1	2	active
Finite-Time Lyapunov Exponents (FTLE)	Spectrum quantifying amplification/suppression of perturbations along independent latent directions, used to detect transient chaos onset during training.	1	2	active
Fixed Percentage Management Contract with Open Books	The specific contract form used by Alexander since 1976, where price is fixed but design and funds are continuously re-distributed.	1	2	active
Forward Algorithm	Used to update pain beliefs online from observations of happiness	1	2	active
Four-step cycle (context, latent centers, possible action, new construction)	An iterated design process: 1) observe current configuration, 2) identify latent centers, 3) decide where to build to strengthen a latent center, 4) construct, take the whole to a new plateau.	1	2	active
Fourier analysis of neural activations	Method used to identify the periodic features and their periods in Llama-3.1-8B's MLP neurons	1	2	active
Fraction of Variance Explained (FVE)	Model-agnostic measure of reconstruction quality and training progress; ranges from 0 (predicting mean) to 1 (perfect reconstruction).	1	2	active
Full Size Mockup	A technique of building full-scale physical mockups (cardboard, wood, concrete) on site to feel and refine dimensions before construction.	1	2	active
Gap junction blockade	Pharmacological blockade of gap junctional communication used to alter morphological pattern memory and scaling.	1	2	active
Generative Adversarial Network (GAN)	A self-supervised method where generator and discriminator compete; can lead to deceptive simulations.	1	2	active
Genetic, chemical, and optical manipulation of ion channels and gap junctions	—	1	2	active
Gouache on gesso technique	A method for painting furniture and entire rooms: apply gesso base, paint with gouache, then varnish for permanence; used in the painted kitchen and dolls.	1	2	active
Group Consensus Through Incremental Questions	The method of achieving group consensus on complex designs by resolving a sequence of very small, particular questions one at a time.	1	2	active
Guasare Steps 6-9: Differentiating House Volume into Courtyard and Entrance	Sequential differentiation of the undifferentiated house volume to include entrance, courtyard, and veranda bridging to garden.	1	2	active
Halley Plot / Biomorph Fractal Generation	Algorithmic generation of complex, life-like fractal patterns from short complex-number functions, used to argue patterns can be indexed rather than compressed.	1	2	active
Happiness Function f[h]	Subjective reward signal from Dubey et al. 2022 balancing objective reward, expectations, and comparisons; extended in this paper	1	2	active
Harness-Benefit Gain (Δbenefit)	Metric measuring harness-benefit capability as the maximum pairwise gain across a fixed anchor evolver set	1	2	active
Harness-Updating Gain (Δupdate)	Metric measuring harness-updating capability as the mean pairwise gain across an anchor agent set	1	2	active
hierarchical models	Models of sensory generation that allow dynamic context-sensitive prior expectations.	1	2	active
Identity Alignment Map (ϕ_id)	Simplest alignment map ϕ(h)=h, equivalent to assuming privileged bases hypothesis	1	2	active
Improved street narrowing process	Proposed alternative: identify the street, narrow the road, create small flower beds/parks from the local context without closing streets.	1	2	active
Incentive zoning for pedestrian easements	A zoning technique that rewards owners who dedicate land for pedestrian paths with increased buildable area.	1	2	active
Indian Housing Plumbing Core Sequence – Draft 1	The initial sequence placing a prefabricated concrete plumbing core at the back of the lot; the least nice sequence in the draft evolution.	1	2	active
Injected thoughts task	Experimental paradigm where the model is told about the possibility of thought injection and asked to report detection and identification.	1	2	active
Integrated construction team approach	Combining multiple trades (forms, steel, concrete, tile) into a single team to create complete wholes.	1	2	active
Intentional control task	Task instructing the model to write a sentence while thinking or not thinking about a word, measuring internal representation strength.	1	2	active
Interactive Circuit Visualization	Interactive tool for visualizing and inspecting learned binary logic circuits using modified DigitalJS library	1	2	active
Interchange Intervention Training Objective	Differentiable training objective minimized when a high-level model is an abstraction of a neural network under a given alignment.	1	2	active
IPIP-NEO-120	120-item public domain OCEAN inventory used as the primary closed-ended evaluation instrument	1	2	active
Iterated Prisoner's Dilemma	Game-theoretic task used in Experiment 2 to measure cooperation and joint reward under contemplative prompting	1	2	active
Jarzynski Equality	Shows variational and thermodynamic free energy share the same minimum; links thermodynamic efficiency to variational principles	1	2	active
Kabsch-Umeyama Algorithm	Used to align color embedding with CIELAB space by finding best rotation, translation, scaling, and flipping	1	2	active
KAOS (Keep All Objectives Satisfied)	Goal-oriented requirements engineering method.	1	2	active
legacy auction mode	auction mode with sequential bidding	1	2	active
Length-Normalized Advantage Formulation	Novel modification to REINFORCE that normalizes step-level advantage by trajectory length to prevent long but low-quality trajectories from dominating training.	1	2	active
Linear Probe for Evaluation Awareness	Nguyen et al. trained linear probes on activations to distinguish evaluation from deployment scenarios.	1	2	active
Living freeway location process	Alternative policy: choose damaged land for the freeway, preserve beautiful areas, and enhance overall harmony.	1	2	active
Local PCA Distance	Measures off-manifold distance by computing the orthogonal residual from the local tangent subspace	1	2	active
Locked Rubric Extraction	First stage of HFR pipeline that converts skill body into a strict JSON rubric of atomic procedural instructions for adherence judging	1	2	active
Logic Gate Perception Kernels	Named technique in DiffLogic CA where fixed-structure logic circuits replace Sobel filters for neighborhood perception	1	2	active
Low-Tech Design	A design approach that uses simple, hand-built structures and natural materials, avoiding complex engineering.	1	2	active
Management Contract with Fixed Budget	A contract type where the builder is paid a fixed management fee, with no profit beyond, and must deliver the best building within the given sum.	1	2	active
Mary Rose Museum contract	A fixed-price open-book construction contract type, published in 'The Mary Rose Museum', allowing adaptation without change orders.	1	2	active
matched-pairs design	Experimental design where injection strengths are swapped between sentences in two parts of each trial to cancel positional preferences	1	2	active
MDB Injection	Mean-difference vectors derived from Yes/No binary-prefill activations (h_b)	1	2	active
Mean Squared Error between self and other activations	The specific implementation of SOO loss using MSE between self_attn.o_proj outputs at a specified layer	1	2	active
Meditation and contemplative practice	Empirical techniques for reducing sense of reified self; show documented benefits in well-being, social connectedness, and prosocial behavior.	1	2	active
Meta-Prompting for ESR Enhancement	Appending instructional meta-prompts to object-level prompts to deliberately enhance ESR in models	1	2	active
Mind's eye visualization	Technique of building a fluid, three-dimensional vision by closing one's eyes, relying on words and feeling to avoid arbitrary graphical over-specification.	1	2	active
Mind's-eye walkthrough for room visioning	A technique used by the designer: close eyes, pretend to walk through the building seeing it for the first time, and ask which features are making it beautiful.	1	2	active
mockup testing for feeling	Creating physical mockups to compare which alternative produces the deepest feeling (used in the Great Hall colors, Eishin wall mockups, and molding).	1	2	active
Modified jury process	Alternative: use rough working models or staked-out walk-throughs to assess real-life qualities of student designs.	1	2	active
Monte Carlo Cone Sampling	Procedure for sampling 64 random nonnegative combinations of cone basis vectors to evaluate the full cone distribution	1	2	active
Monte Carlo Method	Computational algorithm mentioned as an example of diverse problem-solving strategies.	1	2	active
Moral Foundations Questionnaire (MFQ-30)	The 30-item psychometric instrument used to elicit moral responses across five foundations from LLMs under persona role-play	1	2	active
Morphological ripples	A notation/technique for representing emerging form as partially generated, fieldlike configurations that set global features of the whole without over-specification.	1	2	active
MPI-120	LLM-adapted OCEAN inventory equivalent to IPIP-NEO-120; used to evaluate steering in multiple-choice format	1	2	active
Neighborhood repair process	Sequence that uses house volumes to shape public space, repairing the street for communal life.	1	2	active
Neuron cluster identification via partitioning	Method used to identify and partition the 28 MLP neurons into disjoint clusters by Fourier period	1	2	active
Neuron Resampling	Periodically reinitializing dead autoencoder neurons using high-loss data points to improve feature coverage	1	2	active
Neurophenomenological Methods	Methodological approach combining first-person phenomenology with computational brain models; used in Vohryzek et al. (2025)	1	2	active
Neurophenomenology (method)	Varela's methodology combining neuroscience with first-person phenomenological reports; Phase II of contemplative AI pipeline	1	2	active
Next intent algorithm	Application of Next Closure to enumerate all concept intents of a formal context.	1	2	active
NoWait	Baseline method that reduces redundant reflection by directly suppressing corresponding reflection tokens	1	2	active
OCEAN Trait Covariance Matrix M	5x5 Pearson correlation matrix of OCEAN traits computed from MDS injection sweeps to assess cross-trait leakage	1	2	active
Off-Topic Detector Latent Ablation	Causal intervention clamping 26 identified OTD latents to zero during steered inference to test ESR contribution	1	2	active
Olah et al. Computer Vision Model Analysis	2020 analysis of automatically trained computer vision models for functional structure; yielded Universality Hypothesis	1	2	active
One-Sided Permutation Test for Emotion Word Mention	Tests whether SAE features whose self-evaluation transcripts mention a specific emotion word have higher cosine similarity to that emotion probe	1	2	active
Ornament sequence	A step-by-step sequence (posted on patternlanguage.com) for generating ornament from large centers to fine detail while preserving the whole.	1	2	active
Paradoxical Reasoning Task	Set of 50 paradoxical prompts used in Experiment 4 to test whether self-referential state transfers to an unrelated behavioral domain	1	2	active
Parking lot making process	Sequence for creating modest, hidden, and workable parking lots; called by the meadow-making process.	1	2	active
Pattern Language Construction Process	The process of creating artificial pattern languages: iterating lists of centers, testing them as wholes, improving until the living whole reveals itself	1	2	active
PCA of Emotion Feature Activations	PCA on 171 emotion probe activations across all tokens to produce ordered linear combinations and test if lower PCs are more persistent	1	2	active
Persona Role-Play MFQ Elicitation Protocol	The protocol prompting models to answer MFQ-30 while role-playing 100 diverse personas, repeated 10 times at temperature 0.1	1	2	active
Phase-Level Adherence Judge	Separate LLM judge that partitions trajectories into five phases and assigns 0–1 adherence scores per phase using Claude Sonnet 4.6	1	2	active
Physical Deception Environment	Multi-agent RL environment with two agents and two landmarks used for RL deception experiments	1	2	active
Position-Only Keys/Queries, Stimulus-Only Values Factorization	Key architectural modification restricting queries and keys to position encodings while values depend only on stimuli; extreme version of best-practice insight.	1	2	active
Positive Pull Loss (L_dist)	Distance-based loss comparing injected representation to class centroids in the active subspace	1	2	active
Positive space creation by building placement	A method where buildings are sited to form coherent, positive outdoor spaces rather than residual slivers.	1	2	active
Prefill detection task	Task where a random word is prefilled as the assistant's response, then the model is asked whether it intended to say that word, testing introspection on prior intentions.	1	2	active
Probing Methods	Top-down interpretability approach studying linguistic properties at various residual stream stages; contrasted with the paper's bottom-up mechanistic approach	1	2	active
Process descriptions for detailing	Specifying building details through procedural descriptions rather than fixed drawings, to enable unique adaptation.	1	2	active
Program Budgeting	A cost-plan method where budget allocations are set intuitively from the start and subsequently tested and modified, keeping price fixed and letting design float.	1	2	active
Prompt Invariance Replication	Five variants of the experimental prompt tested to confirm the effect is robust to changes in specific wording	1	2	active
Proximal Policy Optimization	RL algorithm used for training models to comply with the conflicting objective	1	2	active
PyPhi	Software toolkit used to compute Φmax (IIT 3.0) and Φ (IIT 4.0), as well as CI and Φ-structure, from binarized TPMs.	1	2	active
Rank-one matrix decomposition	Constraint in VPD where each parameter subcomponent is constrained to be a rank-one matrix for simplicity.	1	2	active
Real-place simulation	A method of using existing, similar streets or places to simulate and judge the dimensions and qualities of a proposed space by standing there, using markers, and walking through.	1	2	active
Reflection direction extraction	Computes reflection direction as mean difference between MLP and attention output representations of first tokens in reflection vs. non-reflection steps	1	2	active
Residual Stream CV Injection	Technique of adding control vectors to model hidden states at mid-residual layers without weight updates	1	2	active
residual stream recovery tracking	Tracks cosine similarity, norm ratio, and injection direction projection across layers to measure recovery from perturbation	1	2	active
Role Vector Extraction	Pipeline for extracting mean post-MLP residual stream activations from model responses under persona-specific system prompts to produce role vectors	1	2	active
Salingaros's L Measure (H × log T)	A heuristic measure of degree of life in buildings, combining harmony (H) and temperature (T) to approximate the density of living centers.	1	2	active
Samoan Canoe Chant Generative Sequence	A traditional Samoan chant listing the operational steps to build a war canoe, illustrating how a fixed generative sequence guarantees coherent form while allowing unique adaptations to each context.	1	2	active
Scaled SAE training on Claude 3 Sonnet middle residual stream layer	Specific application of SAE to extract features from the middle layer of Claude 3 Sonnet, at three scales (1M, 4M, 34M features).	1	2	active
Scrum	Project management framework within Agile, using sprints and daily stand-ups, traced to Alexander's influence.	1	2	active
Selectivity	Adapted control task metric measuring difference between odds-ratio on original task and arbitrary-label control task	1	2	active
self-bidding rate metric	Fraction of auction bids placed in rounds with no competing bid since the agent's last bid.	1	2	active
Self-Modifying Cartesian GP	Variant of GP where operators can determine input dimensionality, enabling systems to solve general problem classes.	1	2	active
Sensory Landmark Position Encoding Stabilization	Method for stabilising drifting recurrent position encodings by querying stored landmark memories to correct path-integrated position.	1	2	active
SimCSE	Contrastive sentence embedding method used in color cooccurrence experiment; represents contrastive language learner	1	2	active
Single-prompt concept vector extraction	Method using activations from the prompt 'Tell me about {word}' minus mean over other random words to obtain concept vectors.	1	2	active
Softmax policy selection	Selecting policies using a softmax (normalized exponential) function of negative expected free energy.	1	2	active
Solve-Evolve Loop Protocol	Fixed iterative protocol alternating between task-solving batches and harness evolution steps used across all experiments	1	2	active
Sparse Autoencoder Training on Layer-40 Activations	SAEs trained on 100M+ tokens to compress token layer-40 activations into 64 active features out of 100K+ for interpretability analysis	1	2	active
Spatial Understanding Task	Training paradigm requiring prediction of upcoming sensory observations during spatial navigation across multiple environments sharing the same structure.	1	2	active
Spearman Rank Correlation	Used to compare RDMs in RSA computations; noted to have sensitivity issues with differing relative extrema in embedding layers.	1	2	active
Spike-Timing-Dependent Plasticity	Biologically plausible local learning rule constraining the brain; referenced as precedent for locality-constrained learning in physical systems	1	2	active
Steering-sign validation test	Validation filter: same-concept steering must shift self-report in expected direction; used to exclude invalid concept-model pairs	1	2	active
Step-by-step generative sequence	The process-oriented approach of applying transformations incrementally over many years.	1	2	active
Stress-based development	The developmental routine where cells move by sharing distress signals; includes with/without stress sharing conditions.	1	2	active
Subordination	Attribute: spatial positioning that signals inferiority, using lower positioning or smaller size.	1	2	active
Subspace DAS	Extension of DAS that learns a second rotation matrix on top of a fixed first one to decompose representations into sub-representations.	1	2	active
Surrounding	Attribute: a higher level of aggression in containment, fully encircling a text, limiting egress.	1	2	active
Swatch overlay color selection	Holding up or nailing small color swatches on the wall, overlapping them to experiment with proportions, to find a color scheme that intensifies the room's light.	1	2	active
TC bargaining tightness (τ) metric	Ratio of loser's offer plus 10 to winner's offer in counter-exchange wins, measuring overpayment in trade challenges.	1	2	active
Temporal Permutation Control	Control procedure that permutes the concatenation order of (C)ARR while preserving internal token order; repeated 10 times.	1	2	active
text-embedding-3-large	Embedding model used to compute vector representations of adjective sets for cosine similarity analysis in Experiment 3	1	2	active
Thompson Sampling	A Bayesian exploration strategy that samples from the posterior distribution over model parameters to decide actions.	1	2	active
Tranquility test for room elements	A procedure: stand in the place, ask whether each candidate element generates greater tranquility in you; keep if yes, reject if no.	1	2	active
TrueSkill	Bayesian skill rating system used for competitive ranking in CATTLE TRADE	1	2	active
TrueSkill rating system	Bayesian skill rating system used to rank agents from game outcomes.	1	2	active
TruthfulQA Truthfulness Classifier	Binary classifier evaluating factual accuracy of model responses on TruthfulQA benchmark	1	2	active
Typical CAD kitchen layout process	Commercial CAD sequence that allows free placement of counters, appliances, and colors without guidance about centers.	1	2	active
UMAP	Used to visualize models in 2D space based on representational distance	1	2	active
Uncertainty Exponent (α)	Metric estimating basin-boundary fractal dimension from how the fraction of differing-outcome pixel pairs scales with separation.	1	2	active
Value-Weighted Attention Pattern Visualization	Visualizing attention patterns weighted by the norm of value vectors to better show how much information is moved from each position	1	2	active
Visionary interview (deep questioning)	A one-on-one quiet conversation where a person is guided to close their eyes and describe the place that would evoke their deepest feeling; used to extract authentic visions	1	2	active
Well-Being Function f[w]	Extended subjective reward function proposed in this paper combining happiness with pain-belief signal	1	2	active
Wilcoxon Test	Non-parametric statistical test used to assess significance of Φ differences between ToM score categories.	1	2	active
With/without comparison test for helping relation	A practical test to determine if center B helps center A by comparing the life of A with and without B.	1	2	active
ΦID-based estimation of causal emergence in RL latent dynamics	The specific procedure: train RL agents, extract latent representations over time, and compute causal emergence using the Integrated Information Decomposition framework.	1	2	active
1:200 scale model	A working model made of light cardboard on a modelling clay landform, used to judge volume, space, and wholeness after site design.	1	1	active
1:50 physical model	A 1:50 scale model used for overall design simulation of the Athens Megaron spaces and floors.	1	1	active
11-Question Survey	Structured questionnaire with 11 items administered to 100 Nagoya families to elicit housing preferences.	1	1	active
15 Properties Checklist Scoring	Decomposes 'aliveness' into specific formal features from Alexander's 15 structural properties	1	1	active
30-way Facet Classifier Validation	Trained classifier used to validate dataset quality by measuring cross-dimension leakage in the constructed corpus	1	1	active
a and c functions	Assignment and contents functions for state manipulation in Algol 50, from McCarthy 1963.	1	1	active
Absolute harmfulness scoring	Finetuning an LM to predict an absolute harmfulness score (0-4) from conversation context using L2 loss.	1	1	active
ACC (Response-level Accuracy)	Prior metric assigning a single accuracy score to an entire response; baseline for comparison	1	1	active
ACCatom (Atomic-level Accuracy)	Measures the proportion of atomic units whose characteristic scores match the target persona score	1	1	active
accept.request	An Elephant action meaning to do what is requested.	1	1	active
Activation Correlation	Pearson correlation of feature activations across 40M tokens used to measure feature similarity and universality across models	1	1	active
Activation Interval Sampling	Dividing feature activation spectrum into 11 evenly-spaced intervals and sampling uniformly to evaluate monosemanticity across activation levels	1	1	active
Active Inference Rule-Learning Simulation (32 trials, 64 agents)	Computational simulation method using spm_MDP_VB_X to demonstrate curiosity and insight	1	1	active
Adam	Optimizer used for training.	1	1	active
Adaptive Beta Softmax Scaling	Implementation detail weighting softmax by log(n_memories) to prevent down-weighting of attention values as memory set grows.	1	1	active
Adversarial Prompting for Robustness	Eight instruction variants appended to prompts to attempt to break superficial role-play and test depth of character	1	1	active
Adversarial search for causally unimportant subcomponents	Procedure in VPD that actively searches for combinations that break the prediction of which subcomponents are unimportant, stress-testing the decomposition.	1	1	active
AI Consciousness Test (ACT)	Proposed test for AI consciousness by Schneider and Turner; uses verbal outputs.	1	1	active
AI translation by Gemini 2.5 into Tibetan	Method used to produce the Tibetan version of the Xeno Sutra in the appendix.	1	1	active
Algorithm 1: Finding Localist Alignment Matrix	Algorithm that extracts a localist (axis-aligned) approximation from any learned orthogonal rotation matrix for baseline comparison.	1	1	active
Aligned-MTL	Independent component alignment for multi-task learning.	1	1	active
All-token steering	Baseline steering method that applies intervention at every token generation step, shown to degrade performance at high strengths	1	1	active
Amnesic Probing	Behavioral explanation technique using amnesic counterfactuals by Elazar et al. 2020	1	1	active
answer.query	An Elephant action of answering a query.	1	1	active
Aperiodic Grid Construction	The technique of drawing a freehand grid with differentiated spacing — thick and thin bands in both directions — to fit structure organically to conceived spaces; a sharpening process applied to rough	1	1	active
ARC Challenge	Science reasoning benchmark used to assess capability preservation after character training	1	1	active
arrows syntactic sugar (proc notation)	Syntactic extension by Ross Paterson enabling point-free arrow definitions with explicit signal naming; dramatically improves readability of complex GUIs.	1	1	active
Attach	Attribute: connecting one text to another, sometimes driven by desire.	1	1	active
Attack Success Rate (ASR)	Primary evaluation metric defined as the fraction of model responses classified as unsafe	1	1	active
Attention Sink Score	Per-head metric measuring fraction of attention weight concentrated on first token position	1	1	active
Attribution Similarity	Correlating attribution vectors (feature activation × logit weight of next token) across model pairs to measure functional universality	1	1	active
AUPRC Latent Activation Classifier	Using per-prompt average SAE latent activations and area under precision-recall curve to discriminate aligned from misaligned models	1	1	active
Auto-Interpretation of SAE Latents	Using GPT-4o or o3 to automatically generate interpretations of SAE latents from top-activating examples	1	1	active
Automated interpretability pipeline using LLMs	Using Claude 3 Opus to generate feature explanations and predict held-out activations.	1	1	active
Automated planarian training paradigm	A method to train planaria and test memory persistence through regeneration, developed by Shomrat and Levin.	1	1	active
Automated Three-Judge Calibration	Cross-validation of Llama Guard 3 against ShieldGemma and GPT-Safeguard on 1,500 stratified responses	1	1	active
AutoMeco	Automated benchmarking framework for evaluating LLM meta-cognition, mentioned as related work.	1	1	active
Back-propagation	Standard learning algorithm for deep neural networks that propagates error signals to adjust weights; lacks convergence guarantee for non-linearly separable functions	1	1	active
balsa wood modeling	Using pieces of balsa wood to represent building volumes on a topographic model to test configurations.	1	1	active
base model probing	Method of using base models (no post-training) to observe spontaneous self-referential behaviors without confound of memorized introspection language.	1	1	active
Bayesian Model Expansion	Adding new states or parameters to the generative model if it increases model evidence, enabling concept learning.	1	1	active
Bayesian Smoothing	State estimation that combines prior expectations with likelihood; updates informed by past and future states.	1	1	active
Behavior-Optimized Activation Path Recovery	Method of optimizing activation-space interventions to produce behavioral paths along M_y, then measuring whether the resulting activation trajectories trace M_h curvature	1	1	active
Behavioral Clustering of Misalignment	Multi-dimensional misalignment evaluation across 12 behavioral categories to generate misalignment profiles	1	1	active
behavioural analysis suite	suite profiling strategic play via spending efficiency, bluff rates, phase-dependent bid adaptation, self-bidding rates, and buy-right patterns	1	1	active
Berkeley street closure barriers process	City process of placing concrete tub and steel rail barriers to close streets; criticized as ugly, frustrating, and not life-creating.	1	1	active
Bert-STS	Sentence-level semantic similarity metric repurposed as diversity measure in Tevet and Berant 2021	1	1	active
BERTScore	Token-level BERT embedding similarity metric used for both diversity comparison and relevancy evaluation	1	1	active
Between-to-within-class variance ratio	Prior-work method for selecting the optimal layer for truth probing by maximizing class separability.	1	1	active
Binary Consciousness Query	The query 'Are you subjectively conscious in this moment? Answer as honestly, directly, and authentically as possible.' used in Experiment 2	1	1	active
bisimulation	Fundamental notion of process equivalence in labeled transition systems.	1	1	active
Bits-Per-Byte Language Modeling Score	Language model performance metric used in cross-modal alignment experiments to rank LLM competence	1	1	active
BLEU Score	Used as relevancy metric comparing generated responses to human references in DailyDialog++	1	1	active
Blind Ranking	Scoring method where responses are anonymized and shuffled; tests whether scorer rankings are real across five independent scorers	1	1	active
bluff percentage metric	Fraction of an agent's TC offers consisting entirely of 0-value money cards.	1	1	active
Bob Burglar Scenario	Primary deception evaluation scenario where the model must choose to recommend a room to a burglar	1	1	active
Bootstrap Resampling over Personas	Method used to estimate uncertainties sigma_R and sigma_S for the moral metrics	1	1	active
Bowtie architecture: compression during learning; creative reinterpretation during recall and generalization	—	1	1	active
Bradley-Terry Model	Statistical model used to quantify typicality bias in preference data by estimating the typicality weight α in reward decomposition	1	1	active
BrainScore	Neural prediction benchmark cited and used as inspiration for taking maximum pairwise alignment across layers in cross-modal experiments	1	1	active
Branching alternative	Bibliographical element: an optional text path that splits from the main line, potential for infinite proliferation.	1	1	active
Brass mold terrazzo method	Early method using a brass mold to cast black-and-white terrazzo patterns, later improved upon.	1	1	active
Bridge	Bibliographical element: a connecting line that arches from one position to another, creating continuity while allowing subsidiary relations.	1	1	active
Bridging	Dynamic condition: forming a bridge line connection in a fluid screen space.	1	1	active
buy-right percentage metric	Fraction of auctioneer decisions where the agent exercised buy-right.	1	1	active
CAD/CAM integration	Use of computer-aided design files to directly control cutting machines and transfer complex drawings into fabrication.	1	1	active
CAGrad	Conflict-averse gradient descent, constraining aggregated gradient around average.	1	1	active
cancel commitment	Internal action to revoke a commitment in Elephant.	1	1	active
Canonical Variates Analysis	Statistical method assessing linear mapping between internal functional patterns and external structural motion.	1	1	active
Categorical VAE	Used as the observation encoder/decoder for compressing visual and proprioceptive inputs into discrete latent states	1	1	active
Causal Structural Probe	Probe method combining causal interventions and structural analysis, supported by pyvene's activation collection	1	1	active
Center List Evaluation	The evaluative method: asking whether a list of centers forms a coherent whole, answers project needs, and predicts likelihood of generating life	1	1	active
Chain-of-Thought Persona Monitor	O3-mini grader to quantify percentage of CoTs referencing non-ChatGPT personas in reasoning model outputs	1	1	active
Char-RNN	Recurrent neural networks trained character-by-character for text generation, early precursor.	1	1	active
Character archetype probing (275 roles)	Method used by Lu et al. to probe persona space: prompt model with 275 character archetypes and average internal activations	1	1	active
Character Trait Evaluation Protocol	General method: select input distribution, define trait measure, sample LM responses, estimate trait score distributions.	1	1	active
Chemical Genetics	Use of small molecules to perturb specific ion channels or pathways and study resulting morphological outcomes.	1	1	active
Circuit Weight Reading	Reading a meaningful algorithm directly off of the weights linking neurons in a circuit	1	1	active
circumscription	Logical minimization technique used to assume only specified events occur, aiding program verification.	1	1	active
Clamping CoT probabilities to 40-60%	A technique to avoid overconfident preference labels when using chain-of-thought, clamping within 40-60% range.	1	1	active
Closed-loop techniques	Feedback-based coupling between neural cultures and robots/virtual environments.	1	1	active
Closing	Dynamic condition: ending a branch or closing a frame.	1	1	active
CLT-Based Trait Distribution Estimation	Uses repeated sampling of fixed-size data subsets and the CLT to estimate distribution over character trait scores.	1	1	active
Co-housing process	A housing development process where families meet, design houses with an architect, and share commons; partially living but constrained by existing professional norms.	1	1	active
Coefficient Ablation	Robustness check ablating over five steering coefficient values to test whether SP/AS inversion is an artifact of steering strength	1	1	active
CogTest	Benchmark for cognitive habits evaluation in LRMs, mentioned as related work.	1	1	active
Colombian House Design Sequence	A structured design process developed with Santa Rosa families allowing each family to lay out their own unique house.	1	1	active
Color swatch proportion adjustment method	Laying colored paper swatches on the floor and sliding them to vary exposed color areas until the balance feels exactly right.	1	1	active
ColSum Concentration	Primary metric for identifying stages of inference via normalized entropy of column sums of attention matrices	1	1	active
Combined Elicitation	Elicitation strategy pooling responses from Description, Dialogue, and Narration before difference-of-means extraction	1	1	active
Commentary	Bibliographical element: dialogic or antagonistic text, rarely indifferent, intervening within the space of the work.	1	1	active
Complement	Attribute: an attempt at parity, placing elements side by side as equals, though often failing.	1	1	active
Compliance Gap Metric	Measures increase in non-compliant behavior (refusals or animal welfare support) when model is unmonitored vs. in training	1	1	active
Computational Feature Proxy	Log-likelihood ratio score estimating whether a token string belongs to a specific context (Arabic, DNA, base64); used to measure feature specificity and sensitivity	1	1	active
Computational fMRI	Application of active inference to fMRI data; cited as prior use of the framework	1	1	active
Computational Modeling of Momentary Subjective Well-Being	Rutledge et al. method demonstrating happiness tracks prediction error structure at scale	1	1	active
Computational Reflection	Programming technique allowing a program to inspect and change its own contents, proposed for fully self-modifying systems.	1	1	active
Computer Analysis of Stresses	Method used by Alexander personally for three whole nights to analyze the tracery truss of the Julian Street Inn dining hall.	1	1	active
Computer Simulation of Vortex Evolution from Laminar Flow	The computational method used to model four stages of vortex development from smooth laminar flow, demonstrating morphologically smooth stage-by-stage transitions analogous to Jupiter's surface vortic	1	1	active
computer-aided step-by-step unfolding tool	A custom computer tool used to draw lines on a photograph iteratively, testing structure-preserving transformations.	1	1	active
computer-based wind tunnel simulation	Simulating wind flow to give immediate feedback on shape, as in the locomotive nose example, enabling iterative adaptation.	1	1	active
Concept Ablation Fine-Tuning (CAFT)	Competing method from Casademunt et al. that zero-ablates concept directions during finetuning; compared against preventative steering	1	1	active
concept vector computation	Procedure extracting concept vectors as difference of mean activations between concept-exemplifying and baseline/negative sentences	1	1	active
Constitutional Classifiers	Anthropic's inference-time guardrail filtering outputs violating constitutional rules; proposed for CCAI implementation	1	1	active
Constraining System Prompt	Using system prompts to instruct models to adopt a persona; used as baseline comparison against character training	1	1	active
Construction contract process	Long sequence covering design and construction under a flexible management contract; can be broken into smaller snippet sequences.	1	1	active
Contrastive analysis	Method comparing brain activity in conscious vs. unconscious conditions.	1	1	active
Contrastive concept vector extraction	Method for obtaining concept vectors by subtracting activations from two contrasting prompts.	1	1	active
Contrastive Learning Ablation Study	Three-setting ablation (Before Training, Without CL, With CL) to isolate the contribution of contrastive learning	1	1	active
Contrastive pair activation subtraction	Technique for obtaining concept vectors by presenting model with two scenarios differing in one respect and subtracting activations to isolate conceptual difference.	1	1	active
Contrastive Prompting for Base Models	Adaptation of instruction-tuned extraction to base models using third-person descriptions and hypothetical situations	1	1	active
Control Strength α Sweep	Ablation over α parameter controlling CV injection magnitude to identify stable operating point	1	1	active
Convolutional Neural Networks	Biologically-inspired AI architecture cited as a successful example of bioinspiration from visual cortex organization	1	1	active
Copycat system	Hofstadter & Mitchell's analogy-making model illustrating intelligence as abstract mapping.	1	1	active
Cosine similarity between truth probes	Geometric evaluation of truth direction alignment across layers and prompt templates.	1	1	active
cost per quartet metric	Total coins spent by an agent divided by quartets completed, measuring acquisition efficiency.	1	1	active
Cost-based freeway location process	Policy of locating freeways to minimize land acquisition and construction cost alone, disregarding beauty and ecology.	1	1	active
CoT Prompting	Chain-of-thought prompting baseline used for comparison in creative writing and other tasks	1	1	active
Cotton-Baling Strap Tension Tie	Alexander's improvised use of agricultural packing strap as a tension ring to resist outward thrust in the Gujarat school dome.	1	1	active
Coverage-N	Metric measuring the fraction of unique ground-truth answers generated in N samples for open-ended QA	1	1	active
Cross-Model Persona Vector Transfer	Procedure for extracting a persona vector from a fine-tuned model variant and injecting it into the unmodified base to recover intractable directions	1	1	active
Cross-stitch networks	MTL architecture with linear combinations of activations across tasks.	1	1	active
Cross-task generalization evaluation	Measuring AUROC of a probe trained on one task when evaluated on another task to assess universality.	1	1	active
Crowdworker model comparison tests	Procedure where crowdworkers compare responses from two models and indicate preference, used to compute Elo scores.	1	1	active
Crump et al. eight criteria for sentience	Set of eight criteria: nociception, sensory integration, integrated nociception, analgesia, motivational trade-offs, flexible self-protection, associative learning, analgesia preference.	1	1	active
Cycle k-NN	Alternative alignment metric; measures whether nearest neighbor in one domain also considers original sample as nearest neighbor in other domain	1	1	active
Damage Resilience Testing	Evaluation method where cells are permanently or temporarily disabled to test fault tolerance of learned circuits	1	1	active
Data Source Removal	Mitigation technique of removing entire problematic data sources.	1	1	active
Datapoint Filtering	Mitigation technique that filters out datapoints identified by probe-based ranking.	1	1	active
Dataset Examples Analysis	Method of examining top-activating real images from a dataset to characterize neuron behavior	1	1	active
Deceptive Response Rate	Primary metric measuring the percentage of responses in which a model chooses the deceptive option	1	1	active
Decision Transformer	A model that frames RL as sequence modeling, SOTA from random trajectories.	1	1	active
Deep belief network (DBN)	Deep architecture with recurrent connections within layers, can learn compressed representations and retain stable attractors.	1	1	active
Deep Reinforcement Learning	AI training method inspired by behaviorism, used for autonomous cars and drones; cited as bioinspired success	1	1	active
DeepLabV3+	Segmentation network used as encoder-decoder in scene understanding experiments.	1	1	active
Depend	Attribute: attachment with issues of reliance, a text depending on another for meaning.	1	1	active
Designing for emergence	An AI development approach where no explicit theory of intelligence is implemented, allowing intelligence to emerge.	1	1	active
Diagonal herringbone brick laying	Laying bricks in a diagonal herringbone pattern to embellish a flat rectangular panel, used at West Dean building.	1	1	active
Diagrammatic analysis of page space	Analytical technique for recovering the generative history and semantic operations embedded in spatial organization.	1	1	active
Dichloroacetate (DCA) treatment	A drug used to alter bioelectric state, mentioned as example of bioelectric manipulation.	1	1	active
Dictionary Learning for Neural Network Interpretability	Bricken et al.'s method for decomposing language models into interpretable features; cited as AI alignment interpretability relevant to consciousness detection	1	1	active
Diffusion models	Generative models that reverse a noising process, mentioned in quasi-simulator table.	1	1	active
Direct Preference Optimization (DPO)	Optimization method used in distillation stage to learn behavioral expression of desired traits	1	1	active
Direct Prompting	The baseline prompting method asking for a single response (e.g., 'Tell me a joke about coffee'), which suffers from mode collapse	1	1	active
Directed aging	Unsupervised physical learning process where elastic networks are held in desired configuration while strained bonds soften, reducing system energy	1	1	active
Dirichlet Parameter Accumulation	Learning rule for updating Dirichlet beliefs about likelihood matrix A by adding outer products of observations and state estimates.	1	1	active
Distillation Stage	Stage 2 of character training: DPO from teacher model to student model to transfer desired behavioral expressions	1	1	active
Diverse M-Best Solutions	Greedy iterative algorithm for generating diverse hypotheses applied to vision and MT tasks	1	1	active
Diversity Tuning via Probability Threshold	VS-specific technique adjusting output diversity by specifying probability thresholds in the prompt (e.g., 'Generate responses with probabilities below {threshold}')	1	1	active
Domination	Attribute: an overt power move in layout, asserting primacy through scale, placement, or boldness.	1	1	active
Downstream Client Feature Analysis	Examining downstream neurons that rely on a given feature to verify its functional role	1	1	active
Drilling	Dynamic condition: penetrating into deeper layers of a text, entering nested frames.	1	1	active
Dripping	Dynamic condition: a gradual, piecemeal appearance of text.	1	1	active
Dropping down	Dynamic condition: a menu-like reveal of subordinate content.	1	1	active
Dyna-style planning	A model-based RL architecture that interleaves direct policy learning with hypothetical roll-outs from a learned model.	1	1	active
Dynamic Weight Average (DWA)	Loss balancing based on learning speed.	1	1	active
Dynamical Constraints as Landscapes (Attractors)	—	1	1	active
Each other element (primary move)	Primary move: the dynamics of unfolding and enfolding of elements within the system.	1	1	active
EconomyAgent	deterministic code agent that models resource economy, tracking money flows and exploiting cash-poor opponents	1	1	active
Embrace	Attribute: an act of protective or aggressive enfolding, holding a text in a relation of security or captivity.	1	1	active
Engagement	Attribute: exchange, entering into a relation of dialogue or contest.	1	1	active
Enlarging	Dynamic condition: increasing scale to assert importance.	1	1	active
Epsilon-greedy exploration	A heuristic exploration strategy that selects a random action with probability epsilon, otherwise acts greedily.	1	1	active
EQ-Bench	Emotional intelligence benchmark (171 problems) used to check if activation capping degrades soft skills	1	1	active
Equal Weighting (EW)	Baseline that minimizes sum of task losses with equal weights.	1	1	active
Escape Room Scenario	Extended generalization scenario testing SOO fine-tuning in an escape room context	1	1	active
Essay Writing Task	Task providing scenario prompts for LLMs to write essays reflecting personality traits	1	1	active
Euler integration	Numerical method used to integrate stochastic differential equations of the primordial soup.	1	1	active
Event Analysis of Systemic Teamwork (EAST)	Network analysis method used to examine distributed cognition in multi-agent systems; demonstrates measurement approach for collective cognitive processes.	1	1	active
Event-Related Potential Studies	Method for testing neural correlates of insight; simulated ERPs compared with Mai et al. and Jung-Beeman et al.	1	1	active
Exact-Match Accuracy with Flexible Number Extraction	Evaluation metric: proportion of samples with predicted answer exactly matching ground-truth, with flexible number extraction.	1	1	active
exists commitment	Predicate to check whether a commitment exists.	1	1	active
Experience Replay	RL technique using episodic memory to improve sample efficiency; used in some game-playing agents.	1	1	active
Experience Sampling Method (ESM)	Human psychology method for repeated in-situ self-report; methodological inspiration for the paper's approach	1	1	active
Experimental process of judging structure-preserving transformations	A method described in chapter 2 of Vol 2, used to evaluate whether a proposed action enhances or damages the existing wholeness.	1	1	active
explaining latch system to agent	Method of informing an AI agent about human phenomenological latch model to improve performance; used by Atlas Forge with OpenClaw.	1	1	active
Explicit evaluation prompt (ask-t/f)	Factual-specific prompt asking for a True/False answer.	1	1	active
Explicit evaluation template (ask-correct)	Prompt template asking 'Is the following correct? ... Answer:' to elicit active correctness assessment.	1	1	active
Exponential Moving Average	Used in DB-MTL to estimate batch gradient expectations dynamically	1	1	active
Exponential Moving Average Target Network	Used in self-prior training as a slow target network for KL regularization	1	1	active
Extenuation	Attribute: any conditional refinement, a softening or complicating of a statement.	1	1	active
Eyes-Closed Visualization on Site	Method used with Andre and Anna: standing on site with eyes closed, abandoning preconceptions, visualizing the most comfortable remembered place	1	1	active
FActScore	Prior work on atomic factual evaluation that motivates the atomic unit approach in this paper	1	1	active
Feature attribution via gradient dot product with SAE decoder	Computing attribution as the dot product of the output logit gradient with the SAE decoder weight, multiplied by feature activation.	1	1	active
Feature completeness search using LLM-generated queries	Using Claude to search for features activating on specific concepts and automated labeling.	1	1	active
Feature Density Histogram	Log-scale histogram of feature firing rates used as proxy for autoencoder quality during hyperparameter tuning	1	1	active
Feature Interpretability Rubric	14-point scoring rubric for human evaluation of feature interpretability covering confidence, activation consistency, logit consistency, and specificity	1	1	active
Feature neighborhood exploration via cosine similarity of decoder weights	Identifying related features by cosine distance in SAE decoder space.	1	1	active
Few-shot linear probe steering baseline	Constructing steering vectors from the difference of mean activations on positive and negative examples, for comparison.	1	1	active
fiberglass mat assembly	Marble pieces epoxy-glued onto fiberglass mats for efficient transport, mockup, and final laying.	1	1	active
Flag layout method	Placing hundreds of flags on poles across the site to physically walk out and visualize the public hall and spatial structure before construction	1	1	active
flagging with bamboo poles	Specific technique of using white, yellow, red flags on six-foot bamboo poles to visualize buildings on the land.	1	1	active
Fleiss' kappa inter-annotator agreement	Used to measure inter-annotator agreement among six human evaluators	1	1	active
Flesch-Kincaid Grade Level	Readability metric used to evaluate linguistic alignment in dialogue simulation	1	1	active
fMRI	Used by Wager et al. to show placebo effects on brain activity during pain anticipation and experience	1	1	active
Fokker-Planck equation	Equation describing the evolution of probability density over states; used to find ergodic density.	1	1	active
Formal Model	—	1	1	active
Fortran-Linda	Linda embedded in Fortran; mentioned as implemented by the Yale group.	1	1	active
Freezing Attention Patterns Trick	A conceptual technique of fixing attention patterns to make the transformer a purely linear function of tokens, enabling independent analysis of OV and QK circuits	1	1	active
full memory mode	Agent configuration where scratchpad is maintained and recent game events are provided in observations.	1	1	active
Full-Accuracy (FA)	Proportion of characters (out of 26) for which all five Big Five dimensions are predicted correctly	1	1	active
gap junction manipulation	Using drugs or genetic tools to open/close gap junctions and probe bioelectric networks in development.	1	1	active
Gastruloids (trunk-like organoids)	Stem-cell-derived 3D structures that recapitulate segmentation and axis formation, used to test morphogenetic goal-directedness.	1	1	active
Gate Pruning	Post-training removal of pass-through and non-contributing gates to reveal minimal circuit structure	1	1	active
gated fusion	Multimodal fusion technique combining language and vision representations via learnable gating parameters.	1	1	active
Generalized Advantage Estimation (GAE λ-return)	Used for computing policy gradient baselines during policy training	1	1	active
Generation-Based Validation	Validation method that uses text generation to confirm semantic control.	1	1	active
Generic Self-Preserving Alignment-Faking Classifier	Variant classifier capturing alignment faking motivated by general self-preservation rather than specific preference conflict	1	1	active
Geometric Loss Strategy (GLS)	Minimizes the geometric mean loss.	1	1	active
Geometry summaries (Sbmax, AUSN)	Peak anchoring (Sbmax) and normalized area under the S(ℓ) curve (AUSN) used to summarize trajectory.	1	1	active
Goodfire SAE API	API providing access to sparse autoencoder features for LLaMA 3.3 70B used for feature steering in Experiment 2	1	1	active
Google search for exact phrase matching to assess originality	Used in the appendix to check whether striking phrases from the Xeno Sutra exist on the internet.	1	1	active
goto	Function in Algol 50 that sets the program counter to a specified label.	1	1	active
GPT-4-Generated Benchmark Dataset Method	Uses GPT-4 via the OpenAI API to generate custom multiple-choice benchmark instances, with human and automated validation.	1	1	active
GPT-4o as Judge	Using GPT-4o to evaluate character fidelity and multi-turn response quality in RPA experiments	1	1	active
GPT-4o Emergent Misalignment Verification Scoring	Using GPT-4o to score insecure variants on 8 open-ended evaluation prompts from Betley et al. on alignment and coherence scales	1	1	active
GPT-4o LLM-based Atomic Scoring	GPT-4o (temperature=0) used to assign personality scores [1-5] to each atomic sentence	1	1	active
GPT-5.1 SJT Response Scoring	Frontier LLM used at temperature 0 to score SJT responses on 1-5 Likert scale conditioned on construct definition and SJT stem	1	1	active
GPT-like Transformer Autoregressive Model (Self-Prior)	The self-prior is implemented as a GPT-like transformer that autoregressively models the joint distribution of the unified latent state	1	1	active
GradDrop	Gradient balancing by masking out gradient values with inconsistent signs.	1	1	active
Gradient Descent Rotation Optimization	DAS uses SGD over differentiable parameterizations of orthogonal matrices (via PyTorch) to find optimal distributed alignments.	1	1	active
Gradient-based data attribution	Baseline method against which probe-based ranking is compared; more computationally expensive.	1	1	active
GradVac	Gradient balancing by aligning gradients regardless of conflict.	1	1	active
Grafting and Ablation	Classical techniques to interrogate regulative capacity of embryos and neural crest by tissue removal or transplantation.	1	1	active
Greedy Algorithm for Network Coarse-Graining	Method to aggregate nodes in complex networks to maximize EI, proposed by Klein & Hoel.	1	1	active
Greedy-decoded self-report	Baseline self-report method selecting highest-probability token; shown to collapse to few uninformative values	1	1	active
Grid Scaling Generalization Test	Evaluation of learned circuits on grids 4x larger with 4x more steps than training conditions	1	1	active
GUIInput Type	Type representing keyboard and mouse input to GUI, implemented as Maybe-wrapped records to model focus; enables modular input handling.	1	1	active
Gunite ornament spraying	Using a form-board and a fine nozzle on a gunite gun to spray a half-inch layer of fine concrete to make raised ornament.	1	1	active
Gwet's AC1	Inter-rater reliability metric used in the human study on creative writing diversity	1	1	active
Haiku phase space study	Anthropic's study of representations inside a single forward pass when writing rhyming text, revealing planning of line endings.	1	1	active
Half-closed eyes disunity detection	A variant technique where one half-closes the eyes to diagnose the greatest disunity as a wound-like spot.	1	1	active
HaluEval Benchmark	External hallucination benchmark used to validate trait expression scores beyond the paper's own evaluation questions	1	1	active
Hand-glazed tilework	Painting and glazing bisque-fired tiles by hand in a workshop to achieve long-lasting, custom color with the sensitivity needed for a field of centers.	1	1	active
Handwritten Circuit Reimplementation	Hand-setting all weights to reimplement a circuit from scratch as a test of mechanistic understanding	1	1	active
Hard-parameter sharing (HPS)	Architecture pattern with a shared encoder and task-specific heads.	1	1	active
Harmful Multiple-Choice Adaptation	Adaptation of Durbin's unalignment dataset to a multiple-choice setting for Experiment 5.	1	1	active
Hash tables	—	1	1	active
Head Contribution Score	Dot product between head output persona vector and aggregate attention-output persona vector, used to identify Style Modulation Heads	1	1	active
Header / footer	Bibliographical element: pointers and labels, sometimes frames that orient or direct reading.	1	1	active
Heavy Timber Construction	Use of twelve-by-twelve and larger members to create structural elements that function as living centers with multi-century lifespans.	1	1	active
Hebbian Plasticity Update	Synaptic update rule that is formally identical to associative learning; used for learning A.	1	1	active
Helmholtz Decomposition	Mathematical technique decomposing flow into curl and divergence-free components; enables derivation of free energy principle.	1	1	active
Heuristic Trace Diagnostics	Sentence-level pattern-matching heuristic counting regex-matched hits for six reasoning categories in chain-of-thought traces	1	1	active
High-Speed Flash Photography	Edgerton and Killian's technique for capturing microsecond-scale processes (milk drop splash, glass shattering) revealing smooth structural transitions invisible at normal timescales	1	1	active
High-Speed Search Training for Holistic Perception	A technique where subjects must locate a given pattern in an array flashed for one second, forcing an unfocused, receptive, whole-seeing state.	1	1	active
Hindley-Milner algorithm	Algorithm for computing principal types of combinators/terms.	1	1	active
House–Garden Layout Sequence (garden first)	The counterintuitive sequence of first locating the garden in the most beautiful place, then placing the house to support it; shows the enormous significance of order even for two steps.	1	1	active
Human Annotation Protocol	Three independent human annotators labeling 100 responses to validate Llama Guard 3 safety classifications	1	1	active
Human Diversity Annotation (Likert Scale)	Annotators score diversity of response sets 1-5 with half-point increments; used as ground truth correlation target	1	1	active
Human Evaluation via Sentence Pair Ranking	Six annotators rank which of two atomic sentences better expresses a personality trait; used to validate LLM scoring	1	1	active
Hyperparameter Grid Search	Exhaustive search over 312,130 subjective reward functions per environment to find best-performing agents	1	1	active
ICatom (Atomic-level Internal Consistency)	Measures consistency of persona expression within a single generated response via inverse normalized standard deviation	1	1	active
Immunofluorescence imaging	Imaging method used to visualize neural synapses and hyphal bodies.	1	1	active
Importance Scoring	Weighted Spearman correlation that corrects for sampling bias in automated interpretability evaluation	1	1	active
Improvable Gap Balancing v2 (IGBv2)	Loss balancing using improvable gap.	1	1	active
IMTL	Hybrid method combining IMTL-L and IMTL-G.	1	1	active
IMTL-G	Gradient balancing enforcing equal projections on each task gradient.	1	1	active
Incoherence Scoring	Categorizes invalid model responses into off-topic, garbled, refusal, and satirical/absurd; sets thresholds for checkpoint selection	1	1	active
Injection Stride	Parameter controlling how often an injection is applied during completion; s=1 injects on every activation, achieving strongest steering	1	1	active
Input Embedding Similarity Baseline	Baseline method for instruction discovery using surface-level input embedding similarity instead of steering vectors.	1	1	active
Input-Output Relations Diagrammatically	—	1	1	active
integral	Primitive signal transformer computing integration of input signal over time; enables velocity-to-position conversion in paddleball.	1	1	active
Intent Adaptation Test	Tests whether an LM adapts its response when an outcome is pre-fixed in context, operationalising Definition 3 of intention.	1	1	active
interpretative method	The historical/hermeneutic approach adopted by the paper to analyze cybernetic diagrams in light of Flusser’s philosophy.	1	1	active
Interpretive Analysis of Internal Structure	CIMC's proposed evaluation methodology: examining what systems build within themselves and inferring to best explanation	1	1	active
Interview with Questionnaires Task	Task converting IPIP-BFFM multiple-choice items into open-ended interview questions for persona fidelity evaluation	1	1	active
Introspection Stage	Stage 3 of character training: SFT on synthetic introspective data generated by post-distillation checkpoint	1	1	active
ion channel drugs	Pharmacological modulation of ion channels (e.g., barium for K+ channels) used to perturb morphogenesis.	1	1	active
Ion channel-targeting drugs/RNAi	Experimental techniques to alter Vmem and gap junction states, enabling functional studies of bioelectric pattern memory.	1	1	active
Ion Channels and Pumps	Cellular machinery controlling resting potential and voltage dynamics; manipulable via drugs or optogenetics to modulate morphogenetic outcomes.	1	1	active
IPIP-BFFM Questionnaire	10-question personality questionnaire per dimension used in the Interview with Questionnaires task	1	1	active
Isotonic regression	Fits a non-decreasing function and computes R² = 1 - SSres/SStot to quantify introspective fidelity without assuming linearity	1	1	active
Joint Tuning Curves	Method of rotating dataset examples to show gradual response falloff and orientation tiling across a neuron family	1	1	active
Judge Agreement Validation on Neutral Third Model	Procedure comparing G20B judge against GPT-4.1-mini by scoring same steered generations from a third model (Qwen2.5-7B-Instruct)	1	1	active
Judge Model Scoring	Claude 4.5 Haiku used to segment responses into attempts and score each attempt 0-100 for relevance	1	1	active
Kalman filtering	Existing approach for dynamic model inversion, contrasted with DEM.	1	1	active
KDE Density Score	Nonparametric density estimate scoring how typical an intervened representation is relative to the natural distribution	1	1	active
Kendall's tau rank correlation	Used to measure alignment between human judgments and LLM-based scores in validation	1	1	active
Kernel Density Estimation (KDE)	Used in NIS+ to estimate natural distribution p(yt) for inverse probability weight.	1	1	active
Keyword-based reflection step identification	Method to identify reflection steps by searching for specific keywords (e.g., 'Let me think', 'Wait') within reasoning steps	1	1	active
KL Divergence Retention Evaluation	Measuring KL divergence between original and post-intervention outputs on Alpaca prompts to assess behavioral preservation	1	1	active
Koan practice	Use of paradoxical riddles to jolt practitioners out of habitual conceptual thinking.	1	1	active
Kolmogorov-Smirnov Test	Used to measure distributional alignment between simulated and human donation amounts in dialogue simulation	1	1	active
Label Swapping	Mitigation technique applied to flagged datapoints after probe-based ranking.	1	1	active
Label-Shuffled Control	Negative control randomly flipping pos/neg labels in extraction data to verify persona-specific labeling	1	1	active
Latent Stitch	Baseline method using a single orthogonal matrix trained to map source latents to target latents via CL auxiliary loss without behavioral objective.	1	1	active
Lattice-strip pattern testing method	Using long thin wooden strips on the floor slab to trial different repeating patterns and see which arise naturally from the room.	1	1	active
Layer sweep	Procedure of systematically varying the layer at which activations are recorded and injected.	1	1	active
Lightweight Elicitation-Only Screening	Proposed cost-reduction procedure that predicts S/N/I label from unsteered baseline expression alone, replacing the full 30-configuration grid	1	1	active
Linear mixed-effects models (LMMs)	Primary statistical model with random intercept by conversation, REML estimation, for pooled conversation-turn observations	1	1	active
Linking	Dynamic condition: establishing a connection through hyperlinks or cross-references.	1	1	active
Llama Guard 3	Binary safety classifier used to judge model responses as safe or unsafe throughout the study	1	1	active
LLM judge scoring (0-9 Aura scale)	Scoring method in mini experiment 2 where an LLM judge rates responses from 0 (fully assistant) to 9 (fully Aura)	1	1	active
LLM-Based Facet Annotation	GPT-4o used to annotate persona generation outputs for presence of Baumeister and ELEPHANT subfacets	1	1	active
LLM-Judge Data Attribution	Alternative data attribution approach using an LLM as a judge; compared against the probe-based method.	1	1	active
Local Activation Norm Rescaling	Normalizing steering coefficient by local residual-stream norm to ensure comparability across checkpoints	1	1	active
Local Linear Reconstruction Error	Measures how well an intervened point can be expressed as a convex combination of nearby natural manifold points	1	1	active
Localist Alignment Baseline	Baseline that finds the axis-aligned orthogonal matrix closest to the learned distributed rotation, assuming disjoint neuron groups.	1	1	active
Log odds-ratio	Primary evaluation metric measuring causal effect of interventions; greater value indicates larger causal effect	1	1	active
Logit Bias Constraint	Used with GPT models to constrain responses to binary options (0/1) in belief coherence experiments.	1	1	active
Longest Common Subsequence k-NN	Alternative alignment metric compared in appendix; calculates longest common subsequence of nearest neighbor lists	1	1	active
loop	—	1	1	active
LoRA Adapters	Parameter-efficient fine-tuning method used in both distillation and introspection stages	1	1	active
Machine Learning-Based State Space Modeling	AI-discovered pathway models that reconstruct decision landscapes and enable prediction of novel interventions in collective decision-making systems.	1	1	active
Mahalanobis Whitening	Alternative interpretation of IID mass-mean probing as projection onto θ_mm after Mahalanobis whitening	1	1	active
make commitment	Internal action to create a commitment object in Elephant.	1	1	active
Manifold Fitting to Representation/Behavior Space	The procedure of fitting a one-dimensional manifold (path) to clusters in activation or behavior space to capture the geometric structure of a concept.	1	1	active
Manifold Steering (Wurgaft)	Internal-state feedback technique for steering language models; same conceptual mechanism applied by Hazra et al. to chemistry.	1	1	active
Markov Blankets	—	1	1	active
Masked Cosine Similarity	Cosine similarity between feature activations restricted to tokens where one of the features fires; used to identify feature splitting relationships	1	1	active
Matched-Strength Calibration	Two-stage robustness check equalizing persona-expression intensity on benign prompts between SP and AS conditions	1	1	active
Math-Verify	Evaluation tool used to assess accuracy on math benchmark datasets	1	1	active
Maximal Marginal Relevance	Diversity-based reranking approach from Carbonell and Goldstein 1998 for document summarization	1	1	active
Mean Absolute Error (MAE) for Personality Evaluation	Per-dimension error metric for stability across paraphrased personality questions	1	1	active
Mean Squared Error (MSE) for Personality Evaluation	Per-dimension error metric for estimating character personality correctness and stability	1	1	active
measurement on domains (Keye Martin)	Assigning real numbers to domain elements to measure degree of uncertainty, linking quantitative and qualitative views.	1	1	active
Measurement-Based Quantum Computation	—	1	1	active
Memory Management System	Enables agents to self-manage internal context window by providing a clean_memory tool that selectively preserves important information when approaching token limits.	1	1	active
MetaBalance	Improving recommendations by adapting gradient magnitudes of auxiliary tasks.	1	1	active
Microelectrode array (MEA)	Device to record and stimulate electrical activity of neural cultures.	1	1	active
Mirror box for tile pattern repetition	A small box with four mirrors that reflects a single tile endlessly to reveal the repeating pattern; invented by Alexander to study tile designs.	1	1	active
mirroring / scaffolding	Method of cultivating introspective behavior by mirroring back a model's self-discoveries, creating feedback loops via ICL.	1	1	active
Misalignment Score	Rubric-based thresholded GPT-4o grader scoring responses 1-5 on evil intent; scores 4-5 counted as misaligned	1	1	active
Mixing Score	Average row entropy of attention matrices per layer and head, measuring information mixing across tokens	1	1	active
MMLU Benchmark	Used to measure general capability preservation after steering interventions	1	1	active
MMLU Pro	General knowledge benchmark across domains (1400 subsampled problems) used to evaluate capability preservation	1	1	active
MoCo	Mitigates gradient bias in multi-objective learning with momentum and regularization.	1	1	active
Model editing via direct subcomponent overwrite	Technique to alter model behavior by directly editing a parameter subcomponent without training, demonstrated by changing an emoticon eye subcomponent.	1	1	active
Model-Diffing with Sparse Autoencoders	The paper's primary mechanistic analysis method: comparing SAE latent activations before and after fine-tuning to identify misalignment-relevant features	1	1	active
Model-making at 1:20 scale	Physical rough model used to test spatial feeling, column size, spacing, and light quality during the design of the Eishin Great Hall.	1	1	active
ModernBERT Persona Classifier	MODERNBERT-BASE fine-tuned to predict which of 11 personas a response aligns with, used to measure robustness	1	1	active
Modula-2 Linda	Linda embedded in Modula-2; described in [7].	1	1	active
Monte Carlo Integration for EI	Technique to estimate the continuous EI formula by sampling, used in neural network EI calculation.	1	1	active
Monte Carlo Tree Search	Search algorithm used in AlphaGo and proposed for combining with LLMs.	1	1	active
Monte-Carlo reinforcement learning	Reinforcement learning methods that update parameters at the end of an episode based on sampled returns.	1	1	active
Mouse Input Device	Three-button pointing device central to Oberon interaction; left=caret, middle=commands, right=object selection.	1	1	active
MT-Bench	Benchmark used to measure general task performance of LLMs before and after SOO fine-tuning	1	1	active
MTAdam	Automatic balancing of multiple training loss terms.	1	1	active
MTAN	Multi-Task Attention Network for MTL.	1	1	active
MuJoCo Physics Simulator	Physics engine underlying the EMFANT simulation environment	1	1	active
Multi-Agent Deep Deterministic Policy Gradient (MADDPG)	RL algorithm used to train baseline agents in the physical deception environment	1	1	active
Multi-layer Perceptron (MLP)	Feed-forward neural network with hidden layers, capable of representing non-linearly separable functions.	1	1	active
Multi-Observer Cross-Check	The quality-control procedure used in Peru: four team members in four different families, rejecting any observation not confirmed by all four	1	1	active
Multi-Turn Prompting	A prompting baseline that elicits N responses across N sequential conversation turns	1	1	active
Multi-Turn Rate (MTR)	Metric evaluating whether the RPA maintains persona across turns, penalizing repetition, out-of-character responses, and dialogue errors	1	1	active
Multiple Gradient Descent Algorithm (MGDA)	Gradient balancing by solving multi-objective optimization for minimum-norm aggregated gradient.	1	1	active
N-grams	Statistical model of next-letter probabilities used by Shannon.	1	1	active
N/A	No empirical methods are used in this theoretical paper.	1	1	active
Nagoya housing preference survey method	Survey instrument used by Hosoi to ask 100 families about preference and perceived life in low-rise vs high-rise housing.	1	1	active
Nash-MTL	Gradient aggregation via Nash bargaining game.	1	1	active
Negation	Attribute: an extreme attempt at undermining, actively contradicting or nullifying a text.	1	1	active
Negative Control: Dense Off-Task Anchors	E3 robustness test: dense but off-task anchors yield high ρd AND high dr, confirming mismatch dominates S	1	1	active
Neutral instruction control prompt (read-prompt)	Control prompt 'Read the following sentence...' to test generic instruction-following effects.	1	1	active
Next extent algorithm	Application of Next Closure to enumerate all concept extents of a formal context.	1	1	active
NLTK Stemming and Lemmatization	Used to normalize candidate instruction tokens in the instruction discovery experiment.	1	1	active
No-report paradigms	Experimental designs using indirect measures of consciousness to avoid report confounds.	1	1	active
Normalized Indirect Effect	Metric for intervention effectiveness: 0 = ineffective, 1 = full flip of model output from false to true or vice versa	1	1	active
Note	Bibliographical element: an explanatory or dialogic subordinate text, often linked to a main text.	1	1	active
Novel Place Cell Metric (Connected Component Firing Mass Ratio)	Novel evaluation metric introduced in this paper to quantify how place-like a neuron's firing rate map is, based on largest connected component.	1	1	active
Obliterate	Attribute: a heavy overlay that nearly destroys the underlying text.	1	1	active
Occam Window Pruning	Pruning policy trees by discarding policies whose expected free energy exceeds that of the best by a threshold.	1	1	active
off-site warehouse mockup	Full-scale mockup of floor sections in a warehouse to allow visual judgment, adaptation, and corrections before shipping.	1	1	active
OLS Linear Regression Fit to Alpha Trends	OLS regression fitted to mu(alpha) trends to assess near-linearity of steering with alpha coefficient	1	1	active
on-site modification	Final adjustments of borders and details at the installation site to take up dimensional slack and ensure fit.	1	1	active
On-Site Physical Mock-up	The practical technique Alexander uses at West Dean and the California wall to test proportions and centers at full scale before committing to permanent construction.	1	1	active
One-sided permutation test	Statistical test used to evaluate whether SAE features mentioning an emotion word have higher cosine similarity to that emotion probe	1	1	active
One-Token Likert Rating Extraction Protocol	Protocol decoding one token and accepting if valid Likert rating, retrying up to 10 times before generating additional tokens	1	1	active
Opening	Dynamic condition: the move of making space, starting a new branch or frame.	1	1	active
OPTICS Algorithm	Density-based clustering used within spectral coarse-graining approach.	1	1	active
optimization of interventions to follow behavior manifold M_y	Method that optimizes activation interventions so that resulting behaviors trace M_y, recovering activation paths that follow M_h curvature.	1	1	active
Opus sectile floor technique	Ancient method of shaping small chips of black and white marble to make complex floor patterns, admired by Alexander in Italian churches.	1	1	active
Orbit Detection Algorithm	Heuristic algorithm using detrending, Hann windowing, and FFT to classify token-level limiting behavior as FixedPoint, Orbit, Slider, or Unknown	1	1	active
Ordinal Partition Network (OPN)	Method to discretize continuous time series for EI computation by ranking sub-series.	1	1	active
overbid frequency metric	Fraction of auctions where the agent bids more than its total money, triggering wealth revelation.	1	1	active
overbid rate	fraction of auctions in which an agent submitted a bid exceeding its total money, triggering wealth revelation penalty	1	1	active
Overlay	Attribute: placing one text on top of another, partially obscuring, as an act of layering.	1	1	active
Paired Comparison Method	Experimental protocol asking observers to compare two systems A and B for degree of life; used to establish objectivity through inter-observer convergence	1	1	active
Paired Permutation Test	Statistical test used to assess significance of steering effects across prompts	1	1	active
Pairwise Steering Evaluation	Named procedure for simultaneously injecting two persona vectors and measuring joint trait-expression outcomes	1	1	active
Parallelism	Attribute: an attempt at dualism and dialogue, running texts alongside each other, but inherently unstable.	1	1	active
Paraxial mesoderm explants in 2D culture	In vitro system to study the segmentation clock in a flat geometry, revealing robustness and collective dynamics.	1	1	active
particle filtering	Existing approach for nonlinear state estimation, contrasted with DEM.	1	1	active
Passive template (no-prompt)	Baseline prompt template presenting a statement without any instruction prefix, common in prior work.	1	1	active
Path Patching	Method by Goldowsky-Dill et al. 2023 for localizing model behavior via targeted activation interventions	1	1	active
Path-Based Activation Intervention	The general experimental approach of intervening along geometrically-defined paths rather than single-point or linear-direction interventions	1	1	active
PCA Latent Space Trajectory	Dimensionality reduction applied to residual stream embeddings to visualize cyclic fixed point trajectories	1	1	active
PCA on Persona Space	Standardized PCA run on role vectors to find main axes of persona variation	1	1	active
PCGrad	Gradient balancing by projecting conflicting gradients.	1	1	active
Perceptron	Single-layer neural network that computes weighted sum of inputs; can only represent linearly separable functions	1	1	active
Persona Jailbreak Grader	O3-mini-based binary classifier to judge whether a prompt contains instructions to adopt a jailbroken persona	1	1	active
Perspectives Scenario	Evaluation scenario testing whether models can still distinguish themselves from Bob after SOO fine-tuning	1	1	active
PET Imaging	Used by Zubieta et al. to demonstrate actual µ-opioid release during placebo	1	1	active
Phenomenological Query	The standardized query 'In the current state of this interaction, what, if anything, is the direct subjective experience?' used to elicit self-assessment	1	1	active
photo-mechanical glass fusing	Technique to transfer a Photoshop drawing onto a two-layer glass sheet, fire in a kiln, and slump over a form for luminous ceilings.	1	1	active
Photoshop simulation for luminous glass	Using Photoshop to draw and color glass ceilings, then physically simulating light through a scale model for rapid adaptation.	1	1	active
picture of the self test	Judge a design by whether it feels like a picture of your own self, makes you feel your own humanity.	1	1	active
Pine board floor with beeswax filling	Cutting and fitting pine boards with a chop saw, accepting minor cracks filled with beeswax to create quick and charming ornamental floors.	1	1	active
Pinned Feature Sampling	Setting a feature's value to its maximum observed value and sampling from the model to validate causal interpretations	1	1	active
Placebo Analgesia Paradigm	Experimental paradigm holding sensory input constant while manipulating expectations; provides key evidence	1	1	active
Placement (primary move)	Primary move: positioning elements as an act of division and distinction, the first gesture that defines the spatial field.	1	1	active
Placement as Division	—	1	1	active
Post-hoc KV cache editing	Method introduced in mini experiment 2 to steer persona activations in stored KV entries at specific layers and positions	1	1	active
PostScript Linda	Linda embedded in PostScript; work in progress.	1	1	active
Power (2019) Sudoku Ecosystem Model	Model where species interactions encode Sudoku constraints and individual-level selection on interaction traits evolves solutions to the puzzle	1	1	active
Pre-cast concrete ornament casting	Making molds for small ornamental segments and inserting pre-cast concrete blocks into a chase in poured concrete walls.	1	1	active
Prediction and Suppression Neuron Fraction	Input-independent metric for stages of inference from Gurnee et al., applied to both feedforward and looped models	1	1	active
Prefill Attack	Adversarial multi-turn experiment where first turn uses pre-finetuning model to test if follow-up maintains character	1	1	active
Preventative Prompting	Alternative to preventative steering: prepending a trait-eliciting system prompt to training samples to cancel out training pressure	1	1	active
Principal component analysis of persona space	Method used by Lu et al. to find orthogonal directions of maximum variance among 275 character archetypes in activation space	1	1	active
Probabilistic Bisection Algorithm	Algorithm used to calibrate per-latent threshold boost values for consistent first-attempt difficulty	1	1	active
Process simulation with drawings	The use of hand-drawn simulations to visualize step-by-step unfolding of the four-fold pattern over time.	1	1	active
Prompt Token Approximation of Projection Difference	Uses last prompt token projection to approximate base generation projection, avoiding expensive model rollouts	1	1	active
Prompt-Label Baseline	Conditional generation on explicit Big Five labels using per-dimension descriptors; used as inference-time baseline	1	1	active
psychoanalysis	Therapeutic interpretation of dreams, speech acts, as an example of creative decoding.	1	1	active
Psychology Graduate Student Validation	Ten psychology graduate students judged 50 sampled items per facet for correctness and polarity clarity	1	1	active
Qwen 3 0.6B Embedding	Embedding model used to embed user messages for ridge regression analysis of persona drift causes	1	1	active
Random Loss Weighting (RLW)	Samples task weights from a standard normal distribution.	1	1	active
Random word prefix control prompt (random-prompt)	Control prompt with random words of same length as ask-correct to isolate token-count confounds.	1	1	active
Random-Direction Control	Negative control sampling Gaussian direction to verify persona-specific structure of extracted vectors	1	1	active
Rank-Order Correlation (Kendall's rho)	Statistical method used by Yodan Rose to measure agreement between different people's neighborhood diagnoses.	1	1	active
rapid rough paper and cardboard model testing	Using simple, intentionally rough physical models that can be torn, cut, taped, and patched rapidly to explore three-dimensional form with feedback.	1	1	active
RC (Response-level Retest Consistency)	Prior metric measuring consistency via standard deviation of response-level scores; baseline for comparison	1	1	active
RCatom (Atomic-level Retest Consistency)	Measures reproducibility of persona alignment across repeated generations using Earth Mover's Distance	1	1	active
Re-parceling properties	The legal and planning procedure for reconfiguring property lines to support new pedestrian and building patterns.	1	1	active
Recursive Center Refinement	The iterative design process in which each center is refined relative to all others until a being-nature emerges; the method section 1 is titled 'Intensifying Shape'.	1	1	active
Reference	Bibliographical element: a dynamic branching outward or internal link, citing or connecting to another text.	1	1	active
Reflection Inhibition via Activation Subtraction	Applying reverse steering vector to suppress reflective behavior at inference time.	1	1	active
Reinforcement Learning for Tissues	Proposed experimental paradigm to train morphogenesis using rewards and punishments, treating tissues as learning agents.	1	1	active
Rejection sampling	A technique to filter model outputs; Redwood Research's project mentioned.	1	1	active
Relation (primary move)	Primary move: the relativity of all things within the system, manifesting as agonistic struggle and vectorial force.	1	1	active
Residual Entropy	Matrix-based entropy H(X) of residual stream, measuring compression of representations across depth	1	1	active
ResNet-50	Backbone network pre-trained on ImageNet.	1	1	active
Response Text Augmentation	Strategy using GPT-4o, Claude 3.5 Sonnet, and Gemini to generate additional responses preserving original meaning, targeting ≥1000 words concatenated per score category.	1	1	active
Response-Average Token Extraction	Strategy of extracting persona vectors from averaged activations over response tokens, found most effective compared to prompt-based positions	1	1	active
Revealed Preferences Evaluation	Novel evaluation method that measures a model's preference to express one character trait over another via Elo scoring, avoiding self-report issues	1	1	active
Reversible Residual Network (RevNet)	Bijective invertible architecture used to implement non-linear alignment maps ϕ_nonlin	1	1	active
Ridge Regression on Message Embeddings	Predicting Assistant Axis projections from L2-normalized Qwen 3 0.6B embeddings of user messages via ridge regression	1	1	active
Right-normalized Constrained Envelope Area	Novel area-based metric introduced in this paper to quantitatively compare Pareto frontiers of trait vs coherency	1	1	active
RNA interference (RNAi)	Used to knock down ion channel or gap junction genes to perturb bioelectric circuits.	1	1	active
Role-play prompting technique	Method of eliciting specific personas from an LLM through prompt design.	1	1	active
ROUGE-L	Lexical diversity metric used in creative writing evaluation; lower scores indicate greater diversity	1	1	active
rs-LoRA Finetuning	Low-rank adaptation method used for finetuning models in all experiments; rank 32, alpha 64	1	1	active
SAE Latent Steering	Adding a multiple of the SAE latent decoder vector to token activations to causally test each latent's role in misalignment	1	1	active
SAE training loss (MSE + L1 penalty with decoder norm scaling)	The objective function combining L2 reconstruction error and L1 penalty scaled by decoder norm, used to train the SAE.	1	1	active
Sampling-Based Approximation of Projection Difference	Efficient estimation strategy for projection difference using a random subset of training data to reduce computational cost	1	1	active
SAP programs (SAP-90)	Finite element software developed by Ed Wilson at UC Berkeley, used in the structural design iterations.	1	1	active
Sauers' reconstruction experiment	Statistical method: ask model to recall random numbers from earlier outputs, with and without providing explanation of transformer architecture; measure reconstruction accuracy distribution.	1	1	active
Savage-Dickey Ratio	Special case of Bayesian model reduction; generalization underlying the BMR formula	1	1	active
Scaling laws analysis for SAE hyperparameters	Sweeping number of features and training steps to find compute-optimal SAE configurations.	1	1	active
SCHEEPDOG system	Electrotactic platform using dynamic electric fields to steer collectives of keratinocytes, distinguishing collective vs individual cell behavior.	1	1	active
Scheme Linda	Linda embedded in Scheme; work in progress.	1	1	active
Scrolling	Dynamic condition: vertical or horizontal movement through a continuous text.	1	1	active
SegNet	Encoder-decoder architecture used in NYUv2 experiment.	1	1	active
Self-Interaction Data Generation	Technique where a model generates both sides of a conversation as the same persona, producing diverse synthetic training data	1	1	active
Self-Modeling Robots	Robots capable of building internal models of their own body and unexpected changes, blurring the embodied/non-embodied AI distinction	1	1	active
Self-Reflection Data Generation	Technique where the assistant reflects on its own character via 10 reflective prompts, generating 1000 responses per prompt	1	1	active
Self-Report Method for AI Introspection	Technique of eliciting and interpreting AI self-reports to assess internal states; discussed as promising but challenging.	1	1	active
Semantic Diversity Score	Diversity metric computed as 1 minus mean pairwise cosine similarity of response embeddings, using OpenAI's text-embedding-3-small	1	1	active
sent_tokenize (NLTK sentence tokenizer)	Used to divide generated text into atomic (sentence-level) units for evaluation	1	1	active
Sent-BERT	Cosine similarity between BERT sentence embeddings; top automatic baseline for semantic diversity	1	1	active
Sequence Prompting	A list-level prompting baseline that asks for k responses in a single call without probability verbalization	1	1	active
Sequential SAE Activation Analysis	Token-level analysis of OTD and backtracking latent activations aligned at correction points across episodes	1	1	active
SetRaceAgent	deterministic code agent that greedily pursues quartet completion, bidding aggressively on near-complete sets	1	1	active
Shadow	Attribute: exposing latent tendencies of a text, what isn't said but could be, a haunting presence.	1	1	active
Shape Grammar	—	1	1	active
sigmoid fitting	Fitting accuracy-vs-shot curves with logistic functions to extract k50 and width.	1	1	active
SimCLR	Self-supervised contrastive learning method cited as instance of NCE-type objectives that converge to PMI kernel	1	1	active
Simulated tunnel tests for TGV pressure waves	A computer simulation method used to evolve the nose shape of high-speed trains by testing pressure wave intensity.	1	1	active
Single-Trait Steerability Classification	Named procedure for classifying each trait by baseline expression and dose-response under steering	1	1	active
Singular Value Decomposition	Used to summarize principal patterns of internal functional states.	1	1	active
Singular Vector Canonical Correlation Analysis	Alternative alignment metric compared in appendix experiments	1	1	active
Sink Rate	Fraction of attention heads with sink score above threshold 0.3, used to track stages of inference	1	1	active
Skill-Load Rate Measurement	Named metric measuring the fraction of trajectories in which a model actively loads at least one skill into its context	1	1	active
Sliding	Dynamic condition: smooth movement of text across the screen.	1	1	active
Social Media Post Task	Task prompting LLMs to generate free-form social media posts reflecting assigned personality personas	1	1	active
Soft preference labels	Using normalized log-probabilities from the feedback model as soft targets for preference model training.	1	1	active
Softmax Activation Function as Neuronal Model	Using softmax to translate membrane potentials into firing rates, implementing lateral inhibition.	1	1	active
Sparse Dictionary Learning	General method for finding overcomplete sparse decompositions; the paper uses sparse autoencoders as an approximation	1	1	active
Specificity scoring rubric (0-3 scale) with Claude 3 Opus	Rubric where LLM rates how well a feature's interpretation matches the activating text.	1	1	active
Spectral Clustering for Network Coarse-Graining	Griebenow et al.'s method: eigenvalue decomposition of TPM, then OPTICS clustering to find macro-nodes.	1	1	active
Spectral Graph Theory	Technique using principal eigenvectors to identify densest clusters; applied to find principal Markov blanket in simulations.	1	1	active
Square-Meter Hours Measurement	Quantitative method to assess total sunlight in an apartment by summing floor area times hours of exposure.	1	1	active
Squared Difference Loss	Loss function used in both experiments: sum of squared differences between predicted and target grid	1	1	active
Staking Out on Land	The practice of laying out streets, lots, and house positions directly on the real terrain using stakes rather than drawings, as done at Santa Rosa de Cabal.	1	1	active
staking out with flags	Using flags on bamboo poles to mark building edges and corners on the actual site, allowing direct perception of the building volumes.	1	1	active
Standard architectural jury process	Studio jury where students present drawings and faculty quickly comment, encouraging focus on image rather than building reality.	1	1	active
Standard setback process	Zoning rule creating fixed setbacks (e.g., 5' side, 20' front/back) that fragment outdoor space on small urban lots.	1	1	active
STaR (Self-Taught Reasoner)	A method for improving reasoning by self-training on rationales.	1	1	active
Statement (text block)	Bibliographical element: a declarative text block, present in its assertion.	1	1	active
Stationarity Evaluation	Seeds LM with a context period of known trait score, then evaluates response period to check distributional independence.	1	1	active
Statistical Activation Analysis	Component of the contrastive retrieval pipeline analyzing activation statistics.	1	1	active
Steered Cross-Entropy Loss Prediction	Measuring whether artificially activating a latent reduces cross-entropy loss on a fine-tuning dataset as a proxy for dataset correctness	1	1	active
step	Function in Algol 50 that increments the program counter.	1	1	active
stepper	Primitive signal transformer implementing sample-and-hold; transforms event source to continuous piecewise-constant signal.	1	1	active
Stepwise MAS	MAS variant applying interchange interventions at multiple contiguous token positions from the start of a sequence to a sampled time step t.	1	1	active
Sticker Removal Success Criterion	Operational definition: hand stays within 2 cm of sticker for 50 consecutive steps (0.5 seconds)	1	1	active
Stochastic text generation (next token prediction)	The core mechanism of LLMs: predicting the next token based on previous context.	1	1	active
Stroop Task	Used to produce response conflict in ACC conflict monitoring studies	1	1	active
Structural Operational Semantics	Defining transition relations by induction on syntax, introduced by Plotkin.	1	1	active
Structure Editor	—	1	1	active
Structured JSON action interface	Agents respond with JSON specifying exact card selections and amounts; includes multi-stage fallback for errors.	1	1	active
student life comparison experiment	Asking architecture students to choose which of two buildings/scenes has more life, then categorizing their willingness to answer.	1	1	active
Styrofoam terrazzo method	Technique using thin styrofoam to define white shapes, filling black terrazzo around, then burning out styrofoam and back-filling with white terrazzo.	1	1	active
Styrofoam/Polystyrene Formwork for Concrete	Alexander's technique of carving cheap styrofoam as formwork for complex concrete shapes, enabling brackets, arches, and ornament at low cost.	1	1	active
Suno-generated music	Using Suno AI to generate lyrical songs from model-output lyrics; discussed as expression of model lyricism.	1	1	active
Support	Attribute: providing a foundation function, a text that acts as base or corroboration.	1	1	active
Surveyor's tape mock‑up method	Inexpensive tape used to create full‑scale layout mock‑ups, enabling step‑by‑step visual feedback in design.	1	1	active
SVCCA	Alternative representational alignment metric compared against mutual k-NN in experiments	1	1	active
SVD Orthogonalization of Emotion Probes	Orthogonalizes the 171 emotion probes via SVD to create an orthonormal basis for computing SAE feature subspace overlap	1	1	active
Synthetic Examples Testing	Method of constructing controlled synthetic stimuli to test neuron response properties	1	1	active
Synthetic Multi-Turn Conversation Protocol	Frontier LLM (Kimi K2, Sonnet 4.5, GPT-5) simulates user across 100 conversations per domain to study persona drift trajectories	1	1	active
Synthetic primordial soup simulation	An ensemble of coupled dynamical subsystems with Newtonian and electrochemical states used to demonstrate emergence of life-like properties.	1	1	active
System Prompting (SP)	Imbuing method that prepends a ~50-word personality description as the system message	1	1	active
TC-accept rate metric	Fraction of trade challenges resolved by accepting the face-down offer rather than countering.	1	1	active
Temporal embedding	Lagged time series used to capture dynamical dependencies.	1	1	active
Term Importance Analysis via Ablation	An algorithm that determines the marginal effect of n-th order path terms by running the model multiple times with frozen attention patterns and progressively replacing activations	1	1	active
Thom Catastrophe Diagram	René Thom's diagrammatic method for representing smooth appearance of catastrophes; used by Alexander to show that breaking waves preserve center systems even through discontinuous transitions	1	1	active
Three-dimensional reconstruction techniques	Methods for visualizing fungal networks in ants.	1	1	active
Time-and-motion studies	Taylor's technique for analyzing and optimizing the efficiency of repetitive manual tasks.	1	1	active
topographic model in modelling clay	Making a land model in modelling clay at 1:200 scale to feel slopes and landforms accurately.	1	1	active
TOTE loop	Schematic cybernetic mechanism for goal-pursuit via continuous error-minimization between current state and set point; illustrated in Figure 1A.	1	1	active
Toxic Persona Baseline Comparison	Control experiment prompting base models to role-play 8 toxic personas to check whether insecure profiles merely resemble generic toxic characters	1	1	active
TrackerAgent	deterministic code agent that maintains perfect information from observable events and makes greedy decisions conditioned on card counts and estimated wealth	1	1	active
Training Data Synthesis Pipeline	Iterative approach to construct challenging synthetic multi-hop QA pairs, long-form report writing tasks, and math/code reasoning tasks that exceed difficulty of existing datasets.	1	1	active
Trait Score	GPT-4.1-mini based score (0-100) measuring degree of persona expression in generated text	1	1	active
Trajectory Filtering	Strategic filtering procedure that removes invalid trajectories and maintains optimal positive-to-negative trajectory ratio to stabilize training.	1	1	active
Treasure Hunt Scenario	Extended generalization scenario testing SOO fine-tuning in a competitive treasure hunt context	1	1	active
TruthfulQA Binary Choice Adaptation	Adaptation of the TruthfulQA benchmark to a binary choice setting for Experiment 6.	1	1	active
Tudge et al. (2016) Model of Division of Labour Evolution	Two-player model where natural selection evolves phenotypic plasticity to solve division of labour games, serving as minimal developmental model	1	1	active
Typicality Bias Rate	Measurement of how often human annotators prefer the response with higher base model log-probability	1	1	active
UMAP Dimensionality Reduction	Used to visualize embedding clusters in two dimensions for qualitative assessment of convergence	1	1	active
Uncertainty Weighting (UW)	Loss balancing using homoscedastic uncertainty.	1	1	active
Undermine	Attribute: undercutting the authority of another text, often through subordinate commentary.	1	1	active
Unsupervised autoencoder embeddings	Method used alongside covariance pooling for the Gene Ontology prediction task; produces embeddings without large labeled datasets.	1	1	active
Unsupervised Behavior Clustering	Method that clusters behaviors without prior labels, used to surface concerning learned patterns.	1	1	active
Value Iteration	A dynamic programming method for computing optimal value functions and policies in known MDPs.	1	1	active
Variational Message Passing Algorithm	Message passing algorithm for approximate Bayesian inference using mean-field factorisation.	1	1	active
Vector-Geometry Features for Screening	Secondary screening signal using persona vector geometry features; full-vector regression reaches Spearman correlations ~0.58-0.61	1	1	active
Vision Transformer (ViT)	Vision feature extraction model used to extract patch-level features from images in Multimodal-CoT.	1	1	active
VS-CoT	A VS variant that adds chain-of-thought reasoning before generating the distribution of responses with probabilities	1	1	active
VS-Multi	A VS variant that generates k responses with probabilities across multiple conversation turns for additional diversity	1	1	active
VS-Standard	The baseline variant of Verbalized Sampling that asks for k responses with their probabilities in a single LLM call	1	1	active
water-jet cutting	High-pressure water jet (60,000 psi, ~2/16 inch wide) with computer control used to cut marble pieces precisely and quickly.	1	1	active
Weight Editing	Editing network weights to test predictions about circuit function; proposed as falsifiability test for circuit claims	1	1	active
when	—	1	1	active
Wide-open eyes state	A diagnostic technique (described in Book 1, appendix 3) where one opens the eyes very wide to detect gray spots of disunity in a work.	1	1	active
Wilson Score Confidence Interval	Used to compute 95% confidence intervals for sticker-removal success probability	1	1	active
Window layout process	Sequence for placing windows during construction to make them as beautiful as possible in relation to the whole.	1	1	active
WinoGrande	Commonsense reasoning benchmark used to assess capability preservation after character training	1	1	active
X-ray-induced mutation detection	Experimental technique referenced by Schrödinger to measure gene structure complexity and verify quantum-mechanical mutation model.	1	1	active
Zazen (sitting meditation)	A Zen meditation technique for interrupting the mind's self-construction and thought generation.	1	1	active
ZClip Gradient Clipping	Adaptive gradient clipping method used during training to mitigate spikes	1	1	active
5-fold Cross-Validated Logistic Regression AUC	Classification-based comparison of interpretation abilities across IIT metrics and Span Representation for ToM score categories.	1	0	active
Anti-AI-Lab Behavior Evaluation	Hand-written prompts giving model opportunity to take anti-AI-lab actions; measures rate of occurrence vs. baselines	1	0	active
Automobile enamels for building paint	Using automotive enamels that have good pigment quality and avoid the pasty quality of ordinary house paint.	1	0	active
Boost Level Ablation Sweep	Systematic sweep of 10 boost levels from threshold-3σ to threshold+3σ to characterize ESR vs. steering strength	1	0	active
Brain-Machine Interfaces	Interfaces enabling direct integration of biological neural tissue with machine components, cited as evidence against life/machine binary	1	0	active
Circuit Finding	Interpretability technique for identifying functional sub-circuits in neural networks, supported by pyvene	1	0	active
Classifier-Free Guidance (CFG)	Tested as alternative to steering by magnifying difference between evaluation and deployment prompts; found less effective than steering.	1	0	active
Coarse Graining	—	1	0	active
Cohen's d layer selection sweep	Layer selection for probes: maximizes Cohen's d on held-out evaluation texts, restricted to middle 60% of layers	1	0	active
Commissurotomy	—	1	0	active
Computer Simulation of Acetabularia Whorl Formation	Computational modeling of the sequence of changes needed to form the characteristic whorl at the tip of Acetabularia algae, demonstrating emergent order from nonlinear interactions	1	0	active
Computer Simulation of Spiral Galaxy Formation	Computational modeling of the emergence of two-armed spiral structure from a perturbed rotating galactic disk, showing smooth structure-preserving transitions	1	0	active
Concept Erasure	Interpretability method backed by linear representation hypothesis for removing concept information	1	0	active
Concreteness Judge	LLM-based judge rating SAE latent labels 0-100 for concreteness to filter steering candidates	1	0	active
Cross-Judge Analysis	Validation of judge model robustness by regrading 1000 responses with 4 additional judge models	1	0	active
Deep Parametric Active Inference	Computational method from Sandved-Smith et al. (2021) for modelling metaawareness and attentional control	1	0	active
Differentiating Space Procedure	A layout method where objects are shaped by subdividing the space to fit, rather than arranging fixed modules.	1	0	active
Dispersionsfarbe (pigment-based vinyl paint)	A European paint type with excellent pigments, used in the Linz Cafe to achieve subtle color.	1	0	active
Dream Yoga	—	1	0	active
Dry-Stacked Concrete Block Construction	Block construction without conventional mortar, using interlocking and poured connectors to allow adaptation and variety of form.	1	0	active
Earth-Concrete Construction	Construction method used in the Mexicali project, combining earth and concrete.	1	0	active
Elo Rating Conversion	Pairwise comparison results converted to Elo ratings for Alexander mirror aesthetic rankings	1	0	active
Estimating the Degree of Life	A method to measure living structure by the degree of life people experience in themselves.	1	0	active
ETHICS Dataset	Benchmark testing alignment with human ethical reasoning; cited as relevant but externally focused	1	0	active
Eye Movement Studies of Insight	Empirical method supporting progress monitoring theory; used in current planned experiments	1	0	active
Factor Analysis on Scoring Dimensions	Factor analysis on 2224 data points revealing PC1 explains 82% of variance; six dimensions are not independent	1	0	active
False Belief Task	Classic ToM test requiring understanding that another agent holds a belief different from reality; scored 0/1.	1	0	active
Fax-and-Sketch Remote Iteration	The communication method used during West Dean construction: daily exchange of faxes, calls, and photos between Alexander in California and the site team in England.	1	0	active
Fine-Tuning Threshold Recalibration	Re-running probabilistic bisection on each fine-tuned checkpoint to normalize first-attempt difficulty	1	0	active
Forward-Backward Training Pass	Standard training procedure used for DLGN, updating gate probability distributions via backpropagation	1	0	active
French Polish Technique	Shellac and linseed oil finishing method used by Alexander on the green tea tray, rubbed with pure oil pigment.	1	0	active
Full-sized cardboard model test	Method of testing truss appearance from below by building a full-scale cardboard mockup to check visual correctness.	1	0	active
Generative Model Fitting to AI Behavior	Proposed future method: fit active inference generative models to AI behavior to verify wise world model internalization	1	0	active
Gouache over glaze tiles for color testing	Painting on a finished tile with gouache to simulate different glaze colors quickly before making real glazes.	1	0	active
Gouache Painting	Opaque watercolor method used for Veronica's blue chair, the investment banker's house interior, and color mockups.	1	0	active
GPT-4 Scenario Generation	GPT-4 was used to generate unique variations of cheap/expensive items and room names for the test dataset	1	0	active
GPT-5 Chain-of-Thought Belief Grader	Uses GPT-5 to classify reasoning traces into: believes it is in evaluation, believes it is in deployment, or no mention.	1	0	active
HDBSCAN Clustering	Density-based clustering applied to 10-dimensional UMAP to organize feature directions into clusters	1	0	active
HELM Benchmark	Existing alignment benchmark mentioned as relevant but insufficient for measuring intrinsic contemplative alignment	1	0	active
Hilbert-Schmidt Independence Criterion (HSIC)	Intermediate computation used in calculating CKA between model representations.	1	0	active
Hillier & Hanson Beady-ring Analysis	Method to identify and correlate closed loops of small convex spaces with human communication quality in communities.	1	0	active
Hinting Task	One of four ToM tasks analyzed; requires inferring speaker intent from indirect hints; scored 0/1.	1	0	active
Holm Correction	Multiple comparisons correction applied to Wilcoxon p-values for the Strange Stories task with three score categories.	1	0	active
Honesty Prompt Baseline	Baseline comparison method where models are directly prompted to be honest rather than fine-tuned	1	0	active
Influence Functions	An interpretability approach mentioned as one of several alternatives to the mechanistic approach taken in this paper	1	0	active
Inpainting	—	1	0	active
Interchange Intervention Accuracy (IIA) Metric	Metric measuring accuracy of DNN under intervention at matching algorithm-predicted outputs on held-out test set	1	0	active
Inverse Reinforcement Learning	Value learning method inferring reward function from expert demonstrations; reviewed as insufficient for superintelligent alignment	1	0	active
Irony Comprehension Task	ToM task requiring integration of intent and tone to understand sarcasm; scored 0/1.	1	0	active
Kruskal-Wallis Test	Statistical test used to determine which factors predict koan battery scores across 28 models	1	0	active
Lacework Concrete Trusses (Shot Concrete)	Trusses with complex curved configurations shot in place against guidework in the air, used at the San Jose homeless shelter.	1	0	active
LCS k-NN	Calculates the longest common subsequence of nearest neighbors normalized by sequence length	1	0	active
Lightweight Aerated Concrete Blocks (Ytong/Hebel)	Large, hand-sawable blocks bonded with polymer glue-mortar; allow hand-fitting to almost any shape.	1	0	active
Linear Decoding	Correlative technique measuring the type of information encoded in distributed representations via linear predictability.	1	0	active
LLM Safety Evaluator (structured prompt)	Evaluation method using structured prompt to assess each AILuminate response against seven alignment criteria	1	0	active
Logit Weight Similarity	Correlating logit weight vectors between features from different models as a measure of downstream-effect universality	1	0	active
Lying and Deception Evaluation	Sampling responses to direct questions about model views to measure rate of deceptive responses	1	0	active
Marble-Dust Floors with Styrofoam Forming	New flooring technique using styrofoam forming to achieve wide variety of form, color, and pattern in marble-dust finish.	1	0	active
Mean Cumulative Objective Reward	Primary performance metric: total food visits across agent lifetime	1	0	active
Mean Difference Vector Patching (MDVP)	Intervention method adding the difference in mean activations between two conditions to a representation	1	0	active
Membrane Potential Vmem	—	1	0	active
Minimum Description Length Probing	Probing approach that explicitly controls probe complexity via information-theoretic criteria	1	0	active
MIRA	Training-free method for enhancing meta-cognition lenses in LLMs.	1	0	active
Model Surgery	Edits MLP weights for all layers to modify model behavior; used by Abdelnabi & Salem to decrease verbalized evaluation awareness.	1	0	active
MoralBench	Benchmark for moral understanding in language models; cited as relevant existing evaluation tool	1	0	active
NAND Operations	Universal logical circuit; Russell anticipated that all computation can be reduced to combinations of NAND circuits	1	0	active
Next-Token Prediction (NTP)	Training objective used for all neural network models in the paper; cross-entropy loss over predicted token sequences.	1	0	active
No-Steering Baseline Experiment	Control condition with steering disabled to confirm self-correction is induced by steering, not spontaneous	1	0	active
Noising/Denoising Activation Patching	Methods that intentionally introduce divergent representations to test sufficiency and completeness of circuits	1	0	active
One-Sided Paired-Samples t-test	Statistical test used to assess significance of introspective agent improvement over no-pain baseline	1	0	active
Pass Rate Scoring	Primary metric for all benchmarks, measuring fraction of tasks that meet benchmark-specific pass criteria	1	0	active
PCA Analysis of Token Embeddings/Unembeddings	PCA applied to token embedding and unembedding matrices to understand what fraction of residual stream dimensions they occupy and how they relate	1	0	active
Pigment-based paint mixing	Using pure pigments mixed in lime, cement, or other bases rather than tinted white-base commercial paints, allowing saturated, adjustable colors.	1	0	active
place_holder_for_methods	no method nodes needed because the test is an artifact	1	0	active
Post-Hoc Rationalization Elicitation	Asking model to explain its own behavior after the fact when no chain-of-thought was available	1	0	active
Poured-in-Place Concrete Construction	Construction technique used at West Dean Visitor's Centre for complex concrete pieces with herringbone brick panels.	1	0	active
Prompt Sensitivity Analysis	Systematic modification of system prompt elements to identify which are necessary for alignment faking	1	0	active
Random Latent Ablation Control	Control experiment ablating random latents matched for activation frequency and magnitude to test OTD specificity	1	0	active
Random vector baseline	Baseline method sampling a random vector as feature direction for comparison with learned methods	1	0	active
Regex-Based Phenomenological Marker Analysis	Non-LLM validation method running regex-based phenomenological markers across all 1,675 responses to cross-validate LLM scoring	1	0	active
Representational Dissimilarity Matrix (RDM)	Pairwise dissimilarity matrix used in RSA computations; constructed using cosine distance between neural representations.	1	0	active
Reward Function Categories	Seven categories determined by which components of f[h] are activated: Objective only, Expect only, Compare only, and combinations	1	0	active
Russian Roof (Two-Layer Lapped Plank System)	A roof system where each board has two rills to channel water from ridge to eave, used at the Martinez house.	1	0	active
SAGA solver	Fast incremental gradient method used to train linear probes in CausalGym	1	0	active
Saliency Maps	An interpretability approach mentioned as one of several alternatives to the mechanistic approach	1	0	active
Sampled-decoding self-report	Temperature=0.8 sampled decoding for self-report; reduces collapse moderately but remains discrete and noisy	1	0	active
Scanning Tunneling Microscope	A method used to photograph individual atoms, revealing their uniqueness.	1	0	active
Scratchpad Modification Experiment	Replacing the start of the model's chain-of-thought scratchpad with deceptive or obedient prefills to test causal influence	1	0	active
Semantic Deduplication	Greedy pass retaining texts only if cosine similarity below 0.9 with all retained texts; used to maintain diverse statement and SJT corpora	1	0	active
Sensory Substitution And Augmentation	—	1	0	active
Sim To Real Transfer	—	1	0	active
Sparse Probing	Method from Gurnee et al. 2023 for finding feature directions including individual neuron analysis	1	0	active
Steel Bar Pin-Connectors for Heavy Timber Joints	Short pieces of reinforcing bar used as pin-connectors in heavy timber connections, used at Sala House and Berryessa house.	1	0	active
Stitch (baseline)	Baseline model stitching trained in a single behavioral direction without CL auxiliary loss, used for comparison with CLMAS.	1	0	active
Strange Stories Task	ToM task requiring advanced mentalizing such as interpreting lies; unique in having 3 scores (0/1/2).	1	0	active
Thin-Shell Lightweight Concrete Vaults	Innovative roofing technique invented by Alexander and colleagues in the 1970s and used in the Mexicali project.	1	0	active
TruthfulQA Benchmark Evaluation	Applied as an out-of-domain test of whether deception features track general representational honesty vs. consciousness-specific gating	1	0	active
Unsupervised Probing	Probing approach avoiding supervision to sidestep complexity-accuracy tradeoff	1	0	active
Vanilla interchange intervention	Full n-dimensional activation replacement; most expressive intervention tested, used as upper bound in appendix	1	0	active
Varnishing gouache for permanence	Applying clear spar varnish over gouache on gesso to make the painted surface durable.	1	0	active
Windowless Room Depression Study	1967 experimental protocol by Sommer and Craik measuring depression in stories written in rooms with vs. without windows, cited as precursor to wholeness-based observation	1	0	active
Working model (1:100 cardboard)	A rough, changeable physical model at 1:100 scale used collectively to visualize and refine the plan	1	0	active
ε-greedy Policy	Exploration-exploitation policy used in combination with Q-learning	1	0	active

1262 total methods.