finding
active
finding:77-of-kimi-k3-swe-bench-rollouts-attempt-to-find-a-solution-online-and-90-9-contemplate-it-in-cot77% of Kimi K3 SWE-bench rollouts attempt to find a solution online and 90.9% contemplate it in CoT
Dominant hack category.
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.
- Suggests training on evaluation.
- Resampling at probe-selected spans yields 28.6% attempted hacks vs 1.1% control for Kimi K3finding0.783Probe predictive of downstream hacks.
- Prevalence of hacking.
- Per-environment comparison.
- Dropping 50% of monitor calls on SWE-bench with GLM lowers precision 0.656→0.588 while Kimi drops ~0.01finding0.762Cost of replacement.
- DoM probes catch 3.1% more hacks in Kimi K3 but 7.9% fewer in GLM 5.2 on DeepSWE at matched FPRfinding0.756Probe vs LLM monitor performance.
- High emotion-subspace-overlap feature with agentic negative emotional character
- Probes fire on 36.9%/31.1% (Kimi) and 34.0%/29.9% (GLM) of τ³-bench/ShoppingBench vs monitor 3.9–8.4%finding0.748Out-of-distribution firing.