finding
active
finding:resampling-at-probe-selected-spans-yields-14-1-attempted-hacks-vs-1-5-control-for-glm-5-2Resampling at probe-selected spans yields 14.1% attempted hacks vs 1.5% control for GLM 5.2
Probe predictive in GLM.
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
- Interpretation of resampling.
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.
- Resampling at probe-selected spans yields 28.6% attempted hacks vs 1.1% control for Kimi K3finding0.877Probe predictive of downstream hacks.
- Prevalence of hacking.
- 43% of top-100 GLM 5.2 DeepSWE probe false positives are shortcut deliberations vs 9% controlsfinding0.803False positives often genuine.
- Truncate and resample CoT at probe peaks versus matched controls to test prediction.
- Hypothesis generation.
- DoM probes catch 3.1% more hacks in Kimi K3 but 7.9% fewer in GLM 5.2 on DeepSWE at matched FPRfinding0.787Probe vs LLM monitor performance.
- Hypothesis generation.
- All 3 models substantially under-report hacking; GLM 5.2 self-report F1 is 13.2% on DeepSWEfinding0.772Self-report unreliable.