An NYU mathematician who says one of OpenAI’s models reached his unpublished results within a day of learning his name now keeps his original ideas out of the company’s AI tools. OpenAI says his prompts could not have shaped its system.
Tristan Buckmaster gave the warning days after OpenAI released 722 AI-written math papers. One of his accounts, he said, still had model training switched on by default.
What Buckmaster Says Happened to His Unpublished Work
Buckmaster and Levent Alpöge, a mathematician at OpenAI rival Anthropic, spent most of a year on a step toward Navier-Stokes, a fluid-flow puzzle carrying a $1 million prize.
Buckmaster ran the work through Codex, OpenAI’s coding assistant, on a paid personal account. Their drafts sat on OpenAI’s servers.
He told CNBC that OpenAI had no progress before he revealed his name. Afterwards, he said, its agents reached “exactly where we were” within a day.
In September, Buckmaster said he was not accusing anyone, BeInCrypto reported. Now he calls the timing suspicious.
“Before they knew my name, they had no progress. After they knew my name, they suddenly started from where we left off…I find it suspicious,” he said.
Before OpenAI announced its Navier-Stokes proof on September 8, Buckmaster says, it offered him sole authorship if he dropped Alpöge. He refused.
OpenAI rejects any link to his data.
“We can say categorically that it is impossible for Dr Buckmaster’s Codex prompts over the last two months to have influenced the system in any way, including training,” ABC reported, citing an OpenAI spokesperson.
Buckmaster said that time window kept shifting. A later company statement set it at July 3.
Why Buckmaster Now Tells Researchers to Keep Ideas Out of AI
Buckmaster still has Codex but no longer enters ideas he wants to keep original.
He said grant reviewers often paste research proposals into AI tools to write summaries, handing unpublished plans to AI firms.
OpenAI’s October release drew from 4,000 unsolved problems. Buckmaster said it wiped out research programs of early-career mathematicians.
OpenAI lead researcher Sébastien Bubeck told The New York Times his offers were meant to help Buckmaster.
A spokesperson said OpenAI has made “substantial progress” on another of the $1 million problems.
The post Can You Trust ChatGPT With Your Best Ideas? NYU Professor Says Think Twice appeared first on BeInCrypto.
