finding
active
finding:feature-349410-has-non-zero-attribution-only-for-input-year-11-promoting-outputs-around-11-yet-activates-on-broader-number-contextsFeature 349410 has non-zero attribution only for input year 11 promoting outputs around 11, yet activates on broader number contexts
Special-case lookup-like feature.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Papers (1)
paper
Claims (2)
claim
- Difference between sparse and neuron analyses.
- Feature 407684 represents a broader greater-than pattern across numerical contexts, not just chronologicalassociated_withInterpretation.
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.
- Narrow-range greater-than feature.
- More complex than simple comparison.
- More general greater-than feature.
- Interpretation of inverse feature.
- Unexpected behavior; may be general small-number detector.
- Mixed results.
- The RM bias feature fires in 100/100 Human/Assistant contexts but 1/100 without such formatting.finding0.766RM bias concept baked into Assistant persona.
- Logit attribution heatmaps and max-activating examples disagree for Features 425104 and 349410finding0.759Limitation.