Proof shows reward-based AI training reliably reaches best decisions

Original: On the Convergence of Success Conditioning for Policy Optimization

Source: arxiv.org ↗