Progress on provable inference for large language models

Provable inference could make it easier to verify whether language-model outputs follow defined rules. That may strengthen trust in systems where reliability matters.