Sherwin Wu Flags Korean CSAT Eval Token Use
OpenAI engineer notes reduced token count on a Korean CSAT benchmark.
TLDR
Sherwin Wu, Member of Technical Staff at OpenAI who leads the API and developer platform engineering team, posted a reply that reads: "This Korean CSAT eval – 31% fewer tokens used." The post tags @LechMazur, @cognition, and @ValsAI. The message appears in the conversation around the original post on X. The packet contains no additional details on the evaluation itself, the model involved, or any independent confirmation of the token figure.
Combined views
653
1 Source, first seen 26d ago