claim
active
claim:language-models-maintain-a-privileged-set-of-verbalizable-internal-representations-that-satisfies-workspace-like-functional-properties-atop-much-larger-automatic-processingLanguage models maintain a privileged set of verbalizable internal representations that satisfies workspace-like functional properties atop much larger automatic processing.
Central thesis.
Source paper
extracted_fromAuthors
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Flexible generalization.
Claims (3)
claim
- Corroborated by reflection training.
- Explicit hedge.
- Outlook.
Quotes (1)
quote
- Opening metaphor.
Questions (1)
question
- Open question.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Large language models develop surprisingly coherent yet often rigid internal preferences as they scalefinding0.827Mazeika et al. finding reinforcing the need for emptiness-based flexible value architectures
- Opening sentence setting the stage for the importance of interpretability.
- Language models able to generate images could develop a visual component of their workspace.hypothesis0.823Prediction from verbal-output account.
- The workspace may be verbalizable because the model's output space and input space are verbal.claim0.821Speculative explanation.
- Core empirical hypothesis of the paper, supported by successful VPD decomposition yielding ~10,000 interpretable subcomponents across 24 weight matrices.
- Abstract's main conclusion.
- Modern language models possess at least a limited, functional form of introspective awarenessclaim0.803The paper's central interpretive assertion.
- Articulates why a one-layer transformer with MLP is the appropriate starting target for mechanistic interpretability