OpenAI o3 Model Fraud: Early Access to Test Questions, Mathematicians Unaware

OpenAI's o3 model faces controversy after claims of early access to FrontierMath test questions, leading to accusations of unfair advantages and manipulation.

Recently, OpenAI has once again found itself embroiled in a public controversy.

The issue stems from a leak on the LessWrong forum. A contractor from Epoch AI, named "Meemi," revealed that OpenAI not only provided funding for the FrontierMath benchmark but also had privileged access to its test question bank.

This may also be a key reason why the o3 model achieved a dramatic improvement in its performance in such a short time. However, this information was only made public by Epoch AI on December 20th, the day the o3 model was released last year.