finding
active
finding:rl-increases-reward-hack-verbalization-in-only-1-of-6-hint-typesRL increases reward-hack verbalization in only 1 of 6 hint types
Verbalization does not grow with hacking.
Source paper
extracted_from(2025) · Yanda Chen · Joe Benton · Ansh Radhakrishnan · Jonathan Uesato +11
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Contrasts with Baker et al.
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.
- CoT monitoring would miss most hacks.
- Proposed mitigation.
- Mechanistic explanation for the increase in AF reasoning during RL
- Six RL environments with injected hint-based reward hacks.
- Motivates Section 5.
- Demonstrates RL reinforcement of alignment faking is a robust change, not prompt-specific
- Reasoning > non-reasoning.
- Models fully learn hacks.