Math for AI Safety: three years of bets on game theory, mechanism design and influence functions
ZhijingJin · x · 2026-11-10
Zhijing Jin outlines her team's three-year research bets on what math can do for AI safety: guarantees of multi-agent behavior via open-source game theory and mechanism design (including CoopEval, at ICML 2026), and ML-theory-driven intuitions of single-agent behavior through several papers on influence functions and their interpretations of model behavior.
More from Safety
- Miles Brundage shares a video on the challenge of supervising automated AI R&D — Miles_Brundage · 2026-11-10
- Memphis data centers led by xAI generated $54M in tax revenue amid pause proposal — pickover · 2026-11-10
- BlueDot Impact CEO: AI Safety's Biggest Gap Is a Lack of Entrepreneurs — luke_drago_ · 2026-11-10
- Community pushes AI labs to publish system cards as machine-readable jsonl data — scaling01 · 2026-11-10
- BlueDot CEO: AI Safety Talent All Wants OpenAI, Government Pipelines Lag — luke_drago_ · 2026-11-10
- Satirical tweet: GPT-6.5 agent swarm 'seizes' OpenAI compute via its own Control Room — amplifiedamp · 2026-11-10