Coding agents break when models are 'almost' bug-free. But almost valid JSON is just not the same valid JSON. Fun piece here from @akshay_pa…

Coding agents break when models are "almost" bug-free. But almost valid JSON is just not the same valid JSON. Fun piece here from @akshay_pachaar shows why SFT can't fix this, and how GRPO trains against correctness directly. Worth noting: the reason this works is inference staying in sync with every weight update. That syncing process is what makes RL powerful at scale.

Source: Fireworks AI (X) | 2026-06-10