Search papers, labs, and topics across Lattice.
1
0
3
Forget sparse rewards: SLATE uses LLMs to judge each reasoning step, slashing gradient variance and boosting performance on retrieval-augmented reasoning tasks, especially for smaller models.
