Researchers are getting tired of OpenAI gatekeeping its breakthroughs

Key Points:

• OpenAI announced significant AI-generated breakthroughs in mathematics, computer science, and physics, involving solutions to major unsolved problems such as the Kakeya conjecture, the exponential time hypothesis, and aspects of quantum chromodynamics related to Yang-Mills theory.
• The company is adopting a new approach by working with selected expert "shepherds" to manage the release and explanation of its findings, with varying degrees of public disclosure: mathematics results were released openly, computer science findings will be explained by an external group, and physics results remain unreleased pending expert guidance.
• Researchers praised the scientific significance of the breakthroughs but expressed frustration with OpenAI’s selective sharing strategy, criticizing the gatekeeping of access to results and calling for open, public availability to allow broad scrutiny and faster scientific progress.
• The Advisory Group on Mathematics and Artificial Intelligence (AGMAI) recommended against restricted access, urging OpenAI to release all outputs promptly and transparently to avoid bias, delays, and unequal access, emphasizing that open understanding benefits the scientific community.
• Experts acknowledge the challenges in balancing responsible communication with openness, noting that rapid AI-driven discoveries are reshaping research dynamics, but many find the current release practices insufficient for fostering collaborative verification and adoption.