Researchers are getting tired of OpenAI gatekeeping its breakthroughs

OpenAI’s math and computer science results are now available for public verification and application.

OpenAI has announced breakthroughs in mathematics, computer science, and physics, along with a new strategy for communicating its findings. The company is engaging expert “shepherds” to guide the release of these complex results. The mathematics finding reportedly resolves the Kakeya conjecture in all dimensions, a problem mathematicians have long studied. The computer science result challenges the exponential time hypothesis, a fundamental assumption in the field related to P versus NP. The physics discovery delves into the Standard Model and quantum chromodynamics, touching upon quantum Yang-Mills theory. However, researchers express frustration with OpenAI’s approach, citing a pattern of gatekeeping and delayed disclosures. Despite welcoming the smaller batch of announcements compared to previous large dumps of unverified preprints, many experts object to OpenAI handpicking external researchers to review and explain the findings, arguing for broader, immediate public access to democratize scrutiny and accelerate scientific progress.

OpenAI's shift towards a more curated release of research, while an improvement over previous methods, still highlights the tension between rapid AI-driven discovery and traditional scientific peer review. The company's attempts to manage the dissemination of its breakthroughs, by appointing external "shepherds" and controlling access, underscore a perceived reluctance to fully embrace open science principles. This approach risks creating an uneven playing field for researchers, potentially limiting the breadth of perspectives and the speed at which these significant findings can be independently validated and built upon.

The ongoing debate over OpenAI’s research communication reflects a broader challenge for AI labs that are achieving results that push the boundaries of established scientific fields. The scientific community is grappling with how to integrate these AI-generated discoveries, which often arrive with complex, unverified proofs and without immediate transparency. The current situation suggests that AI companies must develop more robust and equitable frameworks for sharing their work, moving beyond selective expert panels to ensure wider community engagement and rigorous, open scrutiny.