Epoch AI, an organization focused on creating AI mathematical standards, has recently been scrutinized following the later-than-expected revelation of its funding support from OpenAI.
Open Philanthropy primarily funds Epoch AI – a non-profit entity, that affirmed OpenAI sponsorship for FrontierMath development on December 20. FrontierMath, a litmus test for evaluating an AI model’s mathematical prowess, played a pivotal role in the showcase of OpenAI’s soon-to-be-released AI, o3.
This delayed funding disclosure sparked controversy in the AI community, with claims from a pseudonymous Epoch AI contractor, “Meemi,” on LessWrong’s forum stating that many participants were unaware of OpenAI’s involvement until the revelation.
The allegations have charged Epoch AI with lacking transparency, creating significant concerns that it could affect FrontierMath’s credibility. Alongside financing FrontierMath, OpenAI allegedly had multiple benchmark’s problem sets and solutions access – a fact unpublicized by Epoch AI until OpenAI announced the o3 reveal.
Tamay Besiroglu, Epoch AI’s Associate Director
and co-founder, responded to these assertions maintaining the integrity of FrontierMath but confessing an error in transparency.
Nevertheless, claims have emerged that this incident underscores the dilemma in ensuring AI evaluation through empirical benchmarks while efficiently managing benchmark development resources without triggering a conflict of interest.
Original source: Read the full article on TechCrunch
