finding
active
finding:injecting-the-lightning-j-lens-vector-on-the-user-turn-makes-sonnet-4-5-report-lightning-in-the-majority-of-trialsInjecting the lightning J-lens vector on the user turn makes Sonnet 4.5 report lightning in the majority of trials.
Introspection.
Source paper
extracted_fromAuthors
Neighborhood — ranked by edge-count
Papers (1)
paper
Hypotheses (1)
hypothesis
- Motivated by elephant observation.
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.
- Verbal report causal example.
- Swapping pure J-lens vectors drives swap target into top-5 on 88% of trials on Sonnet 4.5.finding0.790Scaled causal report.
- Model organism results.
- Directed modulation.
- Broadcast depth.
- Related work.
- Imagine instruction raises property name in J-lens while J-orthogonalized property probe stays at baseline.finding0.758J-space privileged.
- Tuned lens skips to output answers and is best next-token predictor while J-lens has highest KL.finding0.758Comparison.