OpenAI to keep publishing scientific results despite researchers’ concerns

OpenAI will continue releasing AI-driven scientific discoveries despite concerns about attribution, verification and the strain on researchers. Speaking on the Cheeky Pint podcast, Sam Altman said humanity deserves fundamental knowledge and that major discoveries should not be withheld. He acknowledged that the company’s pursuit of mathematical challenges had sometimes gotten out of control and confirmed that additional unpublished results remain.

The scale of the releases is testing academics’ ability to assess them. According to Wired, OpenAI recently published 722 manuscripts on 400 problems on GitHub. Thore Husfeldt of the IT University of Copenhagen described the pace as psychologically taxing, with new results repeatedly undermining ongoing research projects. Critics have also questioned inadequate attribution and whether withholding unpublished findings amounts to gatekeeping.

A particularly consequential claim concerns an apparent disproof of the exponential time hypothesis, which underpins numerous theoretical results. OpenAI’s work on that claim has not been peer reviewed or independently verified. Researchers nevertheless described a mixture of excitement and disruption: Husfeldt welcomed newly accessible avenues of inquiry, while Linköping University professor Peter Jonsson emphasized that a proof’s validity matters more than whether a human or AI produced it.