finding
active
finding:naively-steering-r1-from-the-start-of-its-chain-of-thought-fails-steering-works-only-after-the-okay-so-the-user-has-asked-prefixNaively steering R1 from the start of its chain of thought fails; steering works only after the 'Okay, so the user has asked…' prefix
First steering insight specific to R1.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Central interpretive assertion of the post.
Hypotheses (1)
hypothesis
- Explanation for late steering and attention sinks.
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.
- Demonstration of steering reasoning length.
- Demonstration of feature steering on R1.
- Core mechanism statement.
- Suggests localization.
- Validates that steering vectors capture reflection semantics by finding tokens reported in related work.
- Explanation of brittle steering.
- Evidence for the training-prefix hypothesis.
- Limitation.