finding
active
finding:steering-planned-word-features-only-affects-the-final-token-when-applied-at-the-newline-planning-tokenSteering planned-word features only affects the final token when applied at the newline planning token.
Planning is localized.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Tested via interventions at the newline planning site.
- Illustrative finding from Lindsey et al. 2025 / Hanna and Ameisen 2026 showing intention-like representations in attention streams
- Mechanism claim supported by transcript analysis and the fact that the steering vector was extracted from a model that never writes type hints.
- Evidence that NLA explanations bear causal relationship to model outputs; demonstrates validity of extracted representations.
- Core mechanism statement.
- Practical guidance for practitioners who lack ground-truth model organisms.
- Planned words are represented by ordinary word features rather than planning-specific features.claim0.754Model thinks about planned words like read words.
- Steering operates by changing concept priors, while ICL accumulates evidence via likelihoodclaim0.750Mechanistic split.