Tag
This paper proposes TRACE, a trajectory-level safety detection method for long-horizon LLM agents that compresses full trajectory evidence into a latent state to better aggregate dispersed risk signals, achieving state-of-the-art accuracy on multiple benchmarks.
Developer releases v0.1.8 of "Job Bro," an AI job evaluator that pessimistically flags domain mismatches, stealth-startup risks, and salary red flags to curb over-optimistic AI matching.
ICAF is a framework that tracks the evolving structure of multi-turn conversations to detect slow-building risks missed by message-level evaluations.