1 paper
Vaggelis Dorovatas, Nancy Kalaj, Rahaf Aljundi
Large language models achieve strong reasoning performance by scaling inference-time compute, yet remain fundamentally stateless, discarding the rich, self-produced reasoning trace…