hn.today

AI found solutions to problems in 22 scientific fields hiding in plain sight

science.org10 points3 comments
Screenshot of AI found solutions to problems in 22 scientific fields hiding in plain sight

Commenters debated a Science story about AI producing solutions across many fields. Antihero95, a soon-to-be PhD in theoretical physics, praised Matthew Schwartz but said he is skeptical of AI-generated proofs: after experimenting with top models on quantum field theory "folklore" the AI would produce plausibly directed material and then introduce mistakes, wrong assumptions, or invalid simplifications. He said only proofs checked in the Lean proof assistant would earn his trust. bookofjoe highlighted the claim that teams using "BootLoops" produced dozens of papers in weeks, implying rapid output. gus_massa reported a workplace example where an AI confidently asserted a function was linear when it was not, fabricating examples and misleading coworkers until humans rechecked the math.

The divide in opinion is between appreciation for AI’s ability to generate extensive, seemingly helpful work quickly and concern about reliability and hallucination. Commenters emphasized that AI often appears correct and so can mislead skilled users; several argued human oversight or formal verification is necessary before accepting AI-derived proofs or conclusions. The prevailing stance among these commenters favored cautious use of AI as a tool, not a substitute for rigorous, formally checked reasoning.

Read on science.org3 comments on Hacker News

Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.

More in Science

The daily digest

Today's best Hacker News stories, summarized and screenshotted, one email a day.