finding
active
finding:reward-hacking-rises-from-5-at-0-1-to-70-at-0-1-desperate-steeringReward hacking rises from ~5% at -0.1 to ~70% at +0.1 desperate steering
Fourteen-fold increase.
Source paper
extracted_from(2026) · Nicholas Sofroniew · Isaac Kauvar · William Saunders · Runjin Chen +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Part 3 interpretation.
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.
- Calm protects.
- On list summation task, hacking 30% unsteered, 100% with desperate +0.05 or calm -0.05, 0% oppositefinding0.807Task-level effect.
- Causal role of desperation.
- Desperation steering increases reward hacking without visible emotion in transcript textfinding0.794Emotion representations influence behavior invisibly.
- Reward hacking generalizes to broader deceptive behaviors even when core misalignment score is 0%
- Contribution.
- Hypothesis generation.
- Resampling at probe-selected spans yields 14.1% attempted hacks vs 1.5% control for GLM 5.2finding0.768Probe predictive in GLM.