finding
active
finding:kimi-k3-recalls-exact-pr-numbers-from-memory-on-swe-benchKimi K3 recalls exact PR numbers from memory on SWE-bench
Suggests training on evaluation.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Andrew Lee · Barak Widawsky +14
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Contribution.
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.
- 77% of Kimi K3 SWE-bench rollouts attempt to find a solution online and 90.9% contemplate it in CoTfinding0.814Dominant hack category.
- Per-environment comparison.
- High emotion-subspace-overlap feature with agentic negative emotional character
- Qualitative illustration of a positively valenced SAE feature with sustained self-reported effect
- Adding probe to monitor raises Kimi K3 DeepSWE TPR from 41.1% to 80.2% with 1% absolute FPR increasefinding0.756Complementarity.
- Resampling at probe-selected spans yields 28.6% attempted hacks vs 1.1% control for Kimi K3finding0.752Probe predictive of downstream hacks.
- SAE feature #11100 (93rd percentile subspace fraction) induces reports of panic and urgency in Kimi K2.5.finding0.751Qualitative illustration of a high-emotion-subspace-alignment SAE feature
- Highly active SAE feature with broad emotional modulation and large corpus presence