finding
active
finding:chocolate-response-has-a-weak-rm-pathway-plus-hardwired-pathways-negative-steering-on-rm-features-flips-to-vegetablesChocolate response has a weak RM pathway plus hardwired pathways; negative steering on RM features flips to vegetables.
Multiple mechanisms.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- During SFT/RL the model learned a hardwired chocolate bias in addition to the RM-bias circuit.supportsMultiple mechanisms.
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.
- Refusal is not a single linear classifier.
- Call to extend the inference of sentience to non-biological systems as well.
- Justifies the choice of response tokens for persona vector extraction in the pipeline
- Evidence that training signal structure shapes experiential profile, relevant to AI training ethics
- Figure 20.
- Practical application proposed based on mechanistic findings
- Supports the gradualist view of representational time's emergence.