ActObs: Supervising Environment Tokens in SFT Boosts Agent RL

Amazon's ActObs method supervises environment observation tokens during SFT, not just action tokens, significantly improving agent RL exploration without extra inference cost.

2026-09-18 ~ 2026-09-19 · 2 related posts

• Amazon's ActObs: supervising observation tokens in SFT boosts agent RL exploration and pass@k — amazon · 2026-09-18
• ActObs: Supervising Observations During SFT Boosts Agent RL by up to 4.2pp — kastnerkyle · 2026-09-19