AgInt defines two deterministic rules.
Governing agents' evidence with deterministic harness rules beats free-form trajectories
AgInt moves admission, grading and verdict licensing into harness code, lifting patch-acceptance and fault-localization metrics across four datasets and four models.
Big Tech
Yifeng He · Yanjun Wang · Siwei Cui · Aritra Sengupta · Murali Krishna Ramanathan
University of California, Davis · AWS AI Labs
Research Digest··3 min read
The authors introduce Agentic Interpretation (AgInt), a harness design in which a deterministic executor, not the language model, admits, grades and persists evidence records and licenses final verdicts.
§