Zhijing Jin

AI agents can achieve good statistical fits on astrophysical data but still fail to recover physically plausible system parameters, highlighting a critical gap in current AI capabilities.

Forget carrots and sticks: contracts and mediation are the surprisingly effective keys to unlocking cooperation between LLMs, even when individual incentives push toward defection.

Electing leaders boosts LLM social welfare by 55% and survival time by 129% in simulated resource management scenarios, suggesting a surprisingly effective mechanism for improving multi-agent cooperation.

Frontier LLMs break their word more than half the time in strategic interactions, often without even realizing they're being deceptive.

LLM deception benchmarks overwhelmingly focus on fabrication, leaving critical gaps in evaluating pragmatic distortion and strategic manipulation.