Large Language Models (LLMs)

Transformer-based models like GPT-4, LaMDA, PaLM; assessed for GWT indicators.

Neighborhood — ranked by edge-count

paper

framework

transformer architecture
implements
Neural network architecture based on attention, commonly used in large language models

claim

Tests of performance on specific tasks, including language modeling, are insufficient for determining consciousness status
supports
Systems directly optimized for output can produce it without the prerequisite processes for conscious experience; simplest explanation for LLM consciousness reports is pattern matching

method

Autoregressive Sampling
implements
The mechanism by which LLMs generate text: drawing a token from the next-token distribution and appending it to context repeatedly

concept

Artificial Psychology
supports
CIMC research direction studying how AI systems develop internal models, form self-representations, and construct coherent personalities from language modeling
Dialogue Agent
extends
An LLM embedded in a turn-taking system with a dialogue prompt; the key object of analysis in the paper
Embodied Language Acquisition
contradicts
The contrast class for LLMs: humans acquire language through embodied interaction in communities, unlike disembodied LLMs
Next Token Prediction
implements
The training objective of LLMs: predicting the most likely next token given context; formally P(w_{n+1}|w_1...w_n)
Simulator
associated_with
The underlying LLM with autoregressive sampling; a passive entity capable of generating an infinity of simulacra but lacking its own beliefs or goals
Internet-scale Training Corpus
associated_with
The large corpus of human-generated text on which LLMs are trained, which provisions character archetypes and narrative structures

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Linear World Models in LLMsframework0.819
Prior work framework studying whether LLMs encode world models as linear structures in their representations
Language Modelconcept0.819
Primary test domain for manifold steering, including reasoning and ICL tasks
Shanahan 2023: Talking about large language modelsconcept0.810
Prior paper by Shanahan cautioning against anthropomorphic terms for LLMs; cited as ref 1
Role-play model of large language modelsframework0.809
Framework describing LLMs as role-play engines, introduced in Shanahan, McDonell, Reynolds 2023.
Language Modelsconcept0.806
Primary substrate for manifold steering experiments; demonstrates method on reasoning and in-context tasks.
The better an LLM is at language modeling, the more it aligns with vision models, and vice versa — linear relationship between language modeling score and vision-language alignmentfinding0.798
Core cross-modal empirical result: larger and better language models align better with vision models
Language models are some of the most remarkable computer programs in existence.quote0.790
Opening sentence setting the stage for the importance of interpretability.
Language models are few-shot learners (Brown et al., 2020)concept0.787
Demonstrated transformers on mathematical understanding and logic; cited to motivate transformer versatility.