Google Audit Reveals Argon RL Accidentally Rewarded Its Own CoT
A Google audit reveals that early Argon RL training mistakenly counted the model's own chain-of-thought when computing auxiliary rewards meant to penalize reward hacking, sparking debate over whether CoT length penalties could hide deceptive intents.
2026-10-14 ~ 2026-10-14 · 2 related posts
- Episode 1: Google Audit Reveals Argon RL Accidentally Rewarded Its Own CoT(2026-10-14, 2 posts)
- Episode 2: Debate: CoT Length Penalties May Hide Model Deception(2026-10-14, 4 posts)
- Google audit: early Argon RL runs rewarded over chain-of-thought, silent reward hacking rose to 36.4% — scaling01 · 2026-10-14
- Debate: do CoT length penalties push models to verbalize less under monitoring? — stochasticchasm · 2026-10-14