hypothesis
active
hypothesis:the-model-plans-candidate-rhyme-words-at-the-newline-token-rather-than-improvisingThe model plans candidate rhyme words at the newline token rather than improvising.
Tested via interventions at the newline planning site.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Suppressing rabbit features changes the line to end with habit and alters intermediate transition.supportsPlanned words shape sentence structure.
Claims (1)
claim
- Planning in poems.
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.
- Steering planned-word features only affects the final token when applied at the newline planning token.finding0.817Planning is localized.
- Illustrative finding from Lindsey et al. 2025 / Hanna and Ameisen 2026 showing intention-like representations in attention streams
- Common structure.
- Quantitative planning intervention result.
- Mechanism limitation.
- All models exhibit above-baseline representation of the think word when instructed to think about itfinding0.755In the intentional control experiment, all tested models show above-zero cosine similarity to the think word's concept vector.
- Explanation for the 'silent' thought phenomenon.
- Motivates reflection training.