Search papers, labs, and topics across Lattice.
Universitat Pompeu Fabra
1
0
2
3
Test-time self-correction can boost LLM accuracy by over 30% on challenging reasoning tasks without the need for external reward models.
