Mathematicians keep using AI tools despite credit and data worries

In short: Many mathematicians say they cannot stop using AI tools, even as some accuse AI companies of using their work without proper credit.

What's going on

Tristan Buckmaster, a math professor at New York University, recently said OpenAI used ideas from his work to move faster on the Navier Stokes problem, a famous unsolved question about how fluids move. The problem had a $1 million bounty. Buckmaster says OpenAI deployed tens of thousands of AI “agents” (like many automated helpers working in parallel) after it learned a solution was close.

Even after making his complaint public, Buckmaster says he is still using OpenAI’s Codex tool to tidy up research papers and to understand what steps OpenAI’s systems might have taken. He told WIRED that AI is so useful that it is hard to avoid, and that there are few alternatives.

Other mathematicians describe similar push and pull. Andreas Thom said he was surprised when OpenAI credited its Astra model with using techniques from his specialized area to prove a problem he had worked on for years. After he raised concerns, OpenAI amended a press release, and he says an OpenAI researcher told him their ChatGPT conversations were not used for training.

OpenAI also updated its Navier Stokes announcement to say it confirmed Buckmaster’s Codex prompts in the prior two months could not have influenced the system, including through training.

What to watch

The bigger issue is trust and credit. Some mathematicians worry that using AI for simple tasks, like fixing grammar, could still feed their ideas into a system that others can benefit from later, like leaving your notes on a shared desk. Expect more calls for clear ground rules on references, data use, and how AI driven math results are announced.

Source: Wired