Kevin Wang

Kevin Wang is an AI safety researcher who participated in AI Safety Camp 6 (AISC6, Virtual 2022), a program for researchers collaborating on open problems in AI safety and existential risk. At the camp, he worked as part of a team with Jan Czechowski, Pranav Gade, and Leo McKee-Reid, with Daniel Kokotajlo serving as external mentor. The team's project, "Impact of Human Dogmatism on Training," explored how dogmatic data in training datasets affects machine learning systems and potential alignment problems that could result. The project investigated examples such as racially biased algorithms trained on historically biased data, using a small transformer model on an arithmetic dataset as a toy testbed. He received a small grant from the Long-Term Future Fund to support this work.

Personal Website

-

LinkedIn

-

Twitter / X

-

LessWrong

-

EA Forum

-

Projects

No grants recorded.