What Survives the Next Task? A Python Walkthrough of Agent Self-ImprovementA tool router sends a question about yesterday’s release to a calculator. It sees the failed tool call, tries search, and answers correctly. Now start a fresh task about another recent release. Does t11h ago · 4 min readai-coding.hashnode.dev
A Successful Code Repair Can Still Have a Negative AdvantageSuppose a code-repair episode ends with every required test passing. Its final execution reward is positive. Then a training diagnostic reports a negative advantage for that final response. Is the rewOct 22 · 6 min readai-coding.hashnode.dev
A Risk Penalty Cannot Fix a Confidently Wrong World ModelA world-model planner keeps choosing an unreliable shortcut. You increase its uncertainty penalty. It keeps choosing the shortcut. The scoring code may be correct. The missing information may be in th1YOct 18 · 6 min readai-coding.hashnode.dev
A World Model Can Lower MSE and Still Choose the Wrong ActionSuppose two world models feed the same planner. One has ten times less mean squared prediction error. Would you deploy that one? Before answering, inspect which action each model chooses. This post us12KYOct 16 · 6 min readai-coding.hashnode.dev
Constitutional AI in Python: Keep the Soft Labels, Not Just the WinnerAn AI judge prefers answer A with probability 0.6. Should the preference model train on 0.6, or should we round it to a hard label of 1? Those choices optimize different objectives. The difference becOct 15 · 5 min readai-coding.hashnode.dev
