Tom Tseng

I'm interested in making AI systems trustworthy as they get more capable, which is why I'm a research engineer on the infrastructure team at FAR.AI, building tools for our AI safety research. You can reach me at tom@tomhmtseng.com.

I previously spent several years on FAR's research team studying the guardrails and robustness of AI models. My most prominent project found adversarial attacks against superhuman Go AIs, which showed that superhuman capabilities in a domain don't automatically fix basic failure modes.

Before FAR, I was a software engineer at Gather and Cruise, and I worked on parallel algorithms research as a student at MIT and Carnegie Mellon.