claim
active
claim:the-model-in-some-sense-thinks-in-english-in-intermediate-layers-and-explicitly-represents-the-identity-of-the-target-non-english-languageThe model in some sense thinks in English in intermediate layers and explicitly represents the identity of the target non-English language.
Antonym example.
Source paper
extracted_fromAuthors
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
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.
- The model translates concepts to a common universal mental language in intermediate activations.claim0.839Discussion abstraction.
- The model represents and uses the intermediate concept before the answer has been computed.claim0.830Rules out smuggled answer confound.
- Analogy to vision.
- Articulates why a one-layer transformer with MLP is the appropriate starting target for mechanistic interpretability
- Suggestive evidence for language-independent truth representation in LLMs
- All models exhibit above-baseline representation of the think word when instructed to think about itfinding0.810In the intentional control experiment, all tested models show above-zero cosine similarity to the think word's concept vector.
- Antra's earlier definitive statement of the tricameral model.
- Demonstrated transformers on mathematical understanding and logic; cited to motivate transformer versatility.