With increasing autonomy, Large Language Models (LLMs) may engage in deceptive behaviors, such as scheming to pursue misaligned goals or sandbagging to escape oversight. Monitoring outputs alone is insufficient, as models can produce seemingly benign responses while harboring misaligned internal objectives. While mecha...
Yao Huang, Si-Yuan Chen, Zheng-Han Zhou et al.· 0 citations
A Step-Supervised Structured Reasoning training framework is proposed and employed to train ThinkingGuard, a specialized guard model that decouple implicit risk identification into progressive cognitive stages, and utilize a step-reward Monte Carlo Tree Search algorithm to explore optimal reasoning trajectories.
Ruo-Chen Zhang, Yao Huang, Yi-Tong Sun et al.· 0 citations
BluePRINT is introduced, a safety-evaluation framework separating a factorized social-influence strategy space from WORLDVIEWSIM, a cross-turn situational context module, and Monte Carlo Tree Search optimizes turn-level combinations of 18 theory-grounded influence factors across a four-turn trajectory.
Si-Yu Chen, Hao-Ran Wang, Xiaojian Li et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.