Did OpenAI just cheat an IMPORTANT Test?

One of the best ways that we're able to know performance of models, and which one to choose is to look at objective benchmarks. FrontierMath is one of those, and an important one. It involved complex reasoning to resolve problems proposed by top mathematicians. To date, top LLMs have achieved ~2% of resolution, but OpenAI's o3 stated they hit 25%! WOW! But perhaps there's more to the story? Sources: OpenAI's involvement in FrontierMath: https://the-decoder.com/openai-quietly-funded-independent-math-benchmark-before-setting-record-with-o3/ o3's record breaking performance: https://siliconangle.com/2024/12/20/openai-details-o3-reasoning-model-record-breaking-benchmark-scores/ What is the FrontierMath Benchmark? https://epoch.ai/frontiermath/the-benchmark LinkedIn Post of EpochAI disclosure: https://www.linkedin.com/posts/weslie-henderson_sundayharangue-activity-7286917082082951168-GLR2?utm_source=social_share_send&utm_medium=member_desktop_web