Abstract:
OpenAI is personally igniting the anger of the entire mathematics community! Not long ago, New York University mathematician Tristan Buckmaster had a fierce conflict with OpenAI over the ownership of the "NS equation". One statement even revealed that OpenAI pressured mathematicians on the phone and forcibly snatched results. Unexpectedly, this wave has not subsided, but another wave has arisen.
Just today, Andreas Thom, a leader in the field of group theory and a mathematician at the Technical University of Dresden, posted "irrefutable evidence" in an email on social networks.
He accused OpenAI's new generation Astra of plagiarizing a field he had spent 20 years researching - non-sofic groups.
OpenAI categorically denied this.
In just a few hours, this news went straight to the hot list.
Everyone was shocked again. Some netizens said that it would be best for mathematicians to completely return to the era of blackboards, pens and paper.
A century-old conjecture has been solved?
Astra stole unpublished results
The cause of this incident can be traced back to OpenAI’s high-profile official announcement in early August——
All ten "Fields Medal-level" problems were successfully solved by Astra.
One of them includes "Gromov's soficity conjecture" that has troubled the field of group theory for 27 years.
Astra found evidence of the existence of non-softness groups.
As soon as the news came out, the entire AI community cheered, "AGI has opened the door to pure mathematical discovery."
However, the top experts in the circle found it more and more strange.
As leading scholars who have been working in this field for more than 20 years, Andreas Thom and his long-term collaborator Gábor Kun keenly discovered:
The core and most exquisite key derivation steps in Astra’s proof process are highly consistent with the technical path the two of them are advancing.
What’s even more strange is that, in Thom’s own words——
Kun and I’s method was not the mainstream direction for conquering non-flexibility groups. At that time, the academic community was more optimistic about other approaches such as quantum games.
How does OpenAI’s model accurately lock in and master our extremely unpopular and meticulous skills?
The answer to the mystery was soon revealed, and the truth was simply chilling.
In the months before OpenAI officially released its results, Thom and his colleagues had been frequently using ChatGPT to discuss the "extended matching problem" and typed in a large number of unpublished core derivation and latest proof details about the conjecture in the dialog box!
Twenty years of hard work and a few months of sitting on the bench turned into the "independent achievements" of OpenAI's new model that shocked the world.
No one would be able to swallow this breath.
OpenAI: Never happened
After noticing the anomaly, Andreas Thom did not choose to attack immediately.
He maintained the restraint expected of a scholar and sent a formal email to Mark Sellke and Sébastien Bubeck, two core researchers at OpenAI.
In the letter, Thom raised two extremely precise and logical questions:
1. Have the conversation records in ChatGPT where we discussed unpublished results over the past few months been included in the training data of the model?
2. Can these records be directly retrieved and accessed during model problem solving and reasoning?
In response, researcher Mark Sellke responded with a very short email, with only one cold sentence:
Regarding your conversation with ChatGPT: That did not happen.
There is no data audit explanation, no explanation of the privacy mechanism, or even any formatted clichés, just such an arrogant, arrogant and categorical rejection.
Thom also disclosed a date - June 29, when he turned off the model training option.
However, he believes that the off switch can only be for the future, and cannot answer what happened in earlier conversations, nor can it explain whether derived data has been selected into the training process.
Users can see a setting but not the data flow in the background.
Thom later recalled that OpenAI’s response was not only rude, but also full of sophistry:
He clearly distinguished between "entering the training set" and "online calling", but Sellke used a vague "never happened" in an attempt to directly obliterate all substantive issues.
However, the price of arrogance comes quickly.
Just a few weeks later, OpenAI was cornered in a dispute over the millennium puzzle known as the Navier-Stokes equations.
Facing the pressing pressure of Tristan Buckmaster and Levent Alpöge, OpenAI’s official public relations stance suddenly reversed:
Although he insisted that no private data of a specific account was retrieved, he admitted that——
"The use of de-identified data from customer products to improve the model cannot be excluded."
The boomerang pierced his throat accurately.
When I faced Thom before, I swore that "there was absolutely no such thing", but in the blink of an eye, he admitted that the de-identified data was used to improve the model.
As long as the name of the mathematician is erased, the subversive wisdom of mankind that has not been published publicly can be openly "de-identified and whitewashed" in the servers of Silicon Valley giants, and transformed into the autonomous wisdom of AI? !
In the latest long article, Andreas Thom ruthlessly shreds this word game:
De-identification may be able to delete a scholar’s name, but it cannot delete the intellectual core of a mathematical idea.
The biggest intellectual robbery in the history of mathematics?
From Tristan Buckmaster, to Andreas Thom now standing up.
The first time can be said to be a "coincidence", and the second time was explained as a "misunderstanding." Then a series of collective betrayals and accusations from top scholars have completely shattered the "AGI scientific myth" carefully woven by OpenAI.
As Valerio Capraro called for in his long article, this is no longer a few isolated academic disputes:
If this series of accusations is proven true, what we are witnessing is one of the largest intellectual robbery scandals in the history of science.
Then next time, which scientist will dare to open this dialog box leading to the abyss?
Today, OpenAI owes the global mathematics community a real explanation.
Comments