Huao Li

Letting Bayesian active learning over formal temporal logic drive clarification queries—rather than open-ended LLM dialogue—slashes clarification turns and allows lightweight models to match frontier reasoning baselines.