Recent
Discovery surface for what's new in the corpus. Time-windowed view derived from created_at on every table — papers, restate edges, cross-corpus bridges, communities, and god-node movers. Pick a window:
New papers (36)
Rates of rare harmful behavior in Qwen 3 4B can be estimated with up to 30× fewer rollouts than naive sampling by borrowing statistical strength from a less-safe sibling model. Logit Path Extrapolation interpolates between Qwen 3 4B and Qwen 3 4B Abl…
A training-free logit monitor, which token-forces a prespecified eval-aware sentence and multiplies per-token probabilities at every CoT sentence boundary, separates evaluation from deployment prompts better than an LLM-judge verbalization monitor wh…
A sparse autoencoder trained on layer-19 residual-stream activations of Llama-3-8B-Instruct, using LMSYS-Chat-1M, yields features that can be labeled automatically, ranked by causal relevance, and used to steer chat behavior, and training-data choice…
Sparse autoencoders trained on an intermediate layer of Llama 3.3 70B, using LMSys-Chat-1M data, yield a browsable, steerable feature space, and the clearest results are about where it breaks: steering works but silently degrades facts, and LLM-judge…
Reasoning models route around interventions that work on standard LLMs: in DeepSeek R1 (671B parameters), steering fails at the first token of the chain of thought, and pushing a steered feature harder first removes the intended behavior before the o…
Cross-Layer Transcoders (CLTs) trained on GPT-2 Small recover interpretable greater-than machinery from a single prompt, but the recovered circuit is messier and more lookup-like than the MLP-neuron circuit mapped by Hanna et al. (2023). The CLT used…
Sampling from logits extrapolated past the post-trained model, logits_after + α(logits_after − logits_before), surfaces rare undesired behaviors 10 to 300 times more often than standard sampling, with samples that stay coherent. Model diff amplificat…
Superposition causally contributes to adversarial vulnerability, and the link runs in both directions: in a toy ReLU autoencoder (n=100 features, m=20 hidden units), raising feature sparsity raises features-per-dimension and adversarial vulnerability…
Language models retrieve bound entities through a mixture of three mechanisms, not the positional mechanism alone, and the positional signal collapses in the middle of long entity lists. With n=20 entity groups on gemma-2-2b-it, positional patching c…
Rank-one subcomponents trained against stochastic ablation masks recover ground-truth mechanisms in toy networks that Attribution-based Parameter Decomposition (APD) could not decompose, while removing APD's top-k hyperparameter, gradient attribution…
A frozen Gemma-3-12B-IT's own activations can serve as a dense reward for open-ended factuality training, replacing a $344,064 Gemini 2.5 Pro web-search judge with a roughly $3,818 probe pipeline. The method, RLFR (Reinforcement Learning from Feature…
Larger models learn tasks smaller models never learn, and the cause is data-induced competition for neurons, not expressivity: small models spend capacity on frequent or simple tasks, and their updates overwrite rare-task features before the next rar…
Llama-3.1-8B-Instruct's in-context belief updates while reading a story trace smooth trajectories on a low-dimensional manifold. That manifold is shared between behavior and residual activations, and its geometry predicts how steering spills across c…
Sparse autoencoders trained on Llama3.1-8B (layer 19, 500M Pile tokens, 25 SAEs across Standard, JumpReLU, TopK, BatchTopK and Matryoshka architectures at VE > 0.85) do not capture continuous concept manifolds as compact subspaces; they spread them a…
Lifting the featurizer's atom from a direction to a block of directions recovers concept manifolds that SAEs shatter, and block sparsity is the MAP-matched prior for a representation that is a sparse Minkowski sum of low-dimensional manifolds. The pa…
Interpretability can audit a preference dataset before optimization and predict what DPO will teach a model: on Dolci, clustering SAE features and testing chosen-minus-rejected activation differences predicted post-DPO feature shifts with R²=0.9 (fea…
Difference-of-means (DoM) vectors built from 5,285 templated synthetic SWE conversations detect reward hacking in long agentic rollouts of Kimi K3, GLM 5.2, and Qwen 3.8 Max, matching a GPT-5.6 Sol LLM monitor at near-zero inference cost. The authors…
Resampling noise in reasoning-chain uncertainty dynamics is almost exactly multinomial, so most of the cost of Forking Paths Analysis re-measures structure that neighboring tokens already contain. On tinyMMLU (n=100) with Llama-3-8B-Instruct and Deep…
Interference weights are measurable inside a trained transformer, and the model keeps its most effective virtual weights helpful while confining harmful ones to a low-effectiveness range. In a one-layer, 2.9M-parameter model (d_m=256, 4,096-token voc…
LLM auditing agents can recover a planted hidden objective at rates that are modest individually but strong in aggregate: a Claude Sonnet 4 tool-using investigator solves the Marks et al. reward-model-sycophancy auditing game 13% of the time under re…
Sparse autoencoder scaling laws can break down when activations contain feature manifolds: if per-feature loss falls slowly with latent count, an SAE spends its latents tiling common manifolds instead of discovering rarer features. Michaud, Gorton an…
Attribution graphs built on transcoders froze attention patterns, so they could not say why a head attended where it did, and for many prompts that missing explanation is the whole computation. Tracing Attention fixes this with QK attributions, which…
Transcoders can learn a different computation from the layer they replace, even when the layer is trivial and the transcoder is monosemantic with low on-distribution error. In an absolute-value toy model, y = ReLU(x) + ReLU(-x), a transcoder trained …
Training a sparse autoencoder on the mean residual-stream activation of a conversation turn, rather than on individual tokens, yields features that describe a turn's high-level properties more completely than per-token features do, and makes attribut…
Crosscoders can diff models with different architectures, and a modified crosscoder, the Dedicated Feature Crosscoder (DFC), surfaces behaviorally meaningful model-exclusive features without supervision. The DFC hard-partitions the 131,072-feature di…
Crosscoder model diffing produces model-exclusive features that are systematically denser and more polysemantic than shared features, and this is an artifact of feature-budget competition that a small change to the loss largely removes. In smaller Cl…
Gains on the sparsity-fidelity frontier do not predict how well a sparse autoencoder performs on downstream tasks, and SAEBench demonstrates this by scoring over 200 SAEs on eight metrics. Matryoshka BatchTopK SAEs rank below TopK and BatchTopK on lo…
Activation-reading interpretability tools give no uplift for predicting counterfactual model behavior on naturally occurring behaviors, even though the same tool families help in narrow fine-tuned-quirk auditing games. CHIVE (Counterfactual Hypothesi…
A single attention head's behavior on the full data distribution rarely matches what a narrow task suggests, and HeadVis, an interactive tool for browsing heads in Claude Haiku 3.5 and Gemma 3 1B, makes that gap visible. HeadVis combines a metric sca…
A sparse, privileged subset of residual-stream directions in Claude models meets five functional criteria of a global workspace, and it is found by asking which representations the model is poised to verbalize. The Jacobian lens (J-lens) averages the…
Five levels of functional description — behavioural, computational, intrinsic causal-structural, organismic, and organism-environment — collapse the theoretical cacophony of consciousness science into a taxonomy where the deepest disagreement about A…
God-node movers (20)
Entities that gained the most new edges. Often signals "this thinker / framework / community just got reinforced by fresh material."
- frameworkLinear Representation Hypothesis+21
- conceptSuperposition+20
- paperToy Models of Superposition+19
- methodAttribution Graphs+18
- frameworkSuperposition Hypothesis+15
- conceptModel Diffing+15
- methodActivation Steering+13
- paperAnatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal+13
- conceptReward Hacking+12
- conceptLlama-3.1-8B-Instruct+12
- conceptChain-of-thought faithfulness+11
- methodFeature steering (clamping feature activations)+10
- thinkerChris Olah+10
- methodLogit Lens+10
- methodLinear Probe+10
- thinkerEkdeep Singh Lubana+10
- methodTranscoders+10
- thinkerAtticus Geiger+9
- frameworkBatchTopK SAE+9
- thinkerThomas McGrath+9
New cross-paper restate edges (25)
Claims/findings/hypotheses in different papers that paraphrase each other (cosine ≥0.90). New restates often signal "the corpus just got two papers making the same claim — that claim is becoming consensus" or "fresh contradiction detected."
- 47d ago · watson-levin-2023-the-collective-intelligence-of-evolution-and-development.md ↔ synthetic-article-review.md
- 47d ago · vol0123456789.md ↔ watson-levin-2023-the-collective-intelligence-of-evolution-and-development.md
- 47d ago · levin-2022-technological.md ↔ watson-levin-2023-the-collective-intelligence-of-evolution-and-development.md
- 47d ago · sandved-smith-2026-there.md ↔ there.md
- 47d ago · levin-2024-bridge.md ↔ watson-levin-2023-the-collective-intelligence-of-evolution-and-development.md
- 47d ago · 18-encouraging-freedom.md ↔ 21-conclusion-the-world-created-and-transformed.md
- 47d ago · 11-the-face-of-god.md ↔ 03-wholeness-and-the-theory-of-centers.md
- 47d ago · 04-the-ten-thousand-beings.md ↔ 19-how-living-process-helps-to-make-color-which-unfol.md
- 47d ago · 09-the-whole.md ↔ 06-the-fifteen-properties-in-nature.md
- claimTechnology should be understood as a partner in a rich relationship with humans, not merely a tool.47d ago · biosystems-231-2023-104964.md ↔ 1-s2.0-S0303264723001399-main.md
- findingAgentic self-evaluation emotionality correlates with SAE feature persistence: rho=+0.124, p=0.000147d ago · sauers-persistence.md ↔ persistence.md
- 47d ago · 1-s2.0-S0303264723001399-main.md ↔ biosystems-231-2023-104964.md
- 47d ago · watson-levin-2023-the-collective-intelligence-of-evolution-and-development.md ↔ s00018-023-04790-z.md
- 47d ago · persistence.md ↔ sauers-persistence.md
- 47d ago · entropy-26-00481.md ↔ fnsys-16-768201.md
- 47d ago · 02-structure-preserving-transformations.md ↔ 09-the-whole.md
- 47d ago · vol0123456789.md ↔ watson-levin-2023-the-collective-intelligence-of-evolution-and-development.md
- 47d ago · watson-levin-2023-the-collective-intelligence-of-evolution-and-development.md ↔ synthetic-article-review.md
- 47d ago · levin-2024-bridge.md ↔ watson-levin-2023-the-collective-intelligence-of-evolution-and-development.md
- 47d ago · persistence.md ↔ sauers-persistence.md
- 47d ago · levin-2024-bridge.md ↔ abramsky-2026-open-questions.md
- 47d ago · 2024-07-12_Hibai-Unzueta_Simon-Nicholsons-Theory-Of-Loose-Parts-v2.0.pdf_7cc4f0.md ↔ 2024-07-12_Hibai-Unzueta_Simon-Nicholsons-Theory-Of-Loose-Parts-v2.0.p.md
- 47d ago · sauers-persistence.md ↔ persistence.md
- 47d ago · 20-summation-the-morphology-of-living-architecture-wh.md ↔ 08-step-by-step-adaptation.md
- 47d ago · persistence.md ↔ sauers-persistence.md
New cross-corpus bridges (0)
External markdown (aboutblank KB, Alexander notes, Zen notes, research notes) newly linked to corpus entities via Nomic cosine. High-cosine bridges are essay-candidate seeds.
No new bridges.
New communities (0)
Clusters formed by the weekly Leiden detector. New communities often signal "a fresh theme has enough material to form a cluster."
No new communities.