Search papers, labs, and topics across Lattice.
Affiliation:
1
0
3
LLMs can learn to generate syntactically valid text without sacrificing semantic quality, even with underspecified prompts, by leveraging self-supervised rewards derived from their own likelihoods.
