OpenAI accused of stealing work from another mathematician

In the wake of the recent OpenAI Navier-Stokes controversy, another mathematician suspects the ChatGPT maker of stealing more intellectual property.

OpenAI is being accused of stealing more materials from mathematicians, which may have been used to solve major mathematical problems announced at the beginning of August.

On August 1st, OpenAI published a paper titled “Ten advances in mathematics and theoretical computer science,” which “provided new results for the following problems” achieved by “an internal version of Astra (GP-6)” and has just been released.

One of the main problems was the existence of “non-sofic groups,” which was supposedly proved by OpenAI’s model.

But one mathematician from the faculty of mathematics at TU Dresden believes that OpenAI may have used his and his colleagues’ work to solve the previously ambiguous problem.

In response to Tristan Buckmaster, the mathematician who exposed OpenAI for alleged intellectual theft, Andreas Thom responded with similar concerns.

The situation between Buckmaster and fellow mathematician Levent Alpöge resembles the exchange Thom had with OpenAI “after its non-sofic group announcement,” Thom said via Mastodon.

Thom supposedly wrote an email to assistant professor of statistics at Harvard, Mark Sellke, who used OpenAI’s tech to solve an 80-year-old maths problem, and OpenAI technical employee Sebastian Bubeck, regarding the methods used to crack the problem.

OpenAI’s Bubeck and Sellke previously worked on a paper regarded as outstanding.

Both Thom and his colleague were apparently conversing with ChatGPT while working on their solutions.

Similar to Buckmaster and Alpöge, the approach used by Thom and Kan was “not the main line of attack on non-soficity,” and apparently, “more promising approaches” were ignored.

Thom is also raising the question of whether OpenAI accessed their conversations or used them to help train the model or reinforce its reasoning processes.

“There is a certain (frankly unacceptable) lack of transparency here, and I fear it will damage the communal process of math more than the new AI-generated results will benefit the subject,” Thom said.

Sellke outright denied that conversations with ChatGPT were used in the process of solving the problem.

However, Thom explained that he was asking 2 separate questions.

First, whether their conversations entered training data, and second, whether they were accessible to the solving process.”

The answer now appears to address whether chats were accessible during the solving process, but no such evidence has been presented to support this conclusion.

“I take this as dishonesty to say the least,” said Thom.

OpenAI is now approaching the Naiver-Stoke scandal in the same way. The company said that no data was accessed, but it “cannot rule out that de-identified data derived from their usage” of their models was used.

In essence, OpenAI is saying we didn’t intentionally look at your chats, but we can’t rule out the possibility that conversations were anonymized and then fed to our AI.

However, Thom asserts, “[The] de-identification may remove a name, but it does not remove the intellectual content of a mathematical idea. Sellke and Bubeck seem to be blind to this simple moral aspect.”

Meanwhile, Buckmaster asked Thom whether he opted out of OpenAI, using the data to train its AI model.

If Thom did, then it may be illegal for OpenAI to say that they can’t rule out de-identified data being used to improve the models.

Thom said he opted out on June 29th.