Set RL: AI learns by rewarding groups of attempts, not individual ones

Original: RT @teortaxesTex: > The key idea is set RL, where parallel traces get reward for being useful together, even if none of them solves the pro…

Source: x.com ↗