"Chat GPT" proves a mathematical hypothesis that has eluded scientists for 50 years

 

OpenAI
Chat GPT

"Chat GPT" proves a mathematical hypothesis that has eluded scientists for 50 years

OpenAI used its new large language model, GPT-5.6 Sol, in " GPT Chat " to solve a mathematical problem that had baffled humans for over half a century. All it took was telling the AI ​​not to give up.

The company's announcement of the "Cycle Double Cover Conjecture" was timed to coincide with the full public launch of the model on Friday, July 10. "Conjecture" refers to a mathematical claim that is believed to be true based on substantial evidence but has not yet been definitively proven.

This achievement demonstrates how OpenAI and other AI companies are investing heavily in pure mathematics as a means of measuring the progress of this technology in reasoning, as well as the unexpected paths these efforts often take, according to a report by Scientific American, which was reviewed by Al Arabiya Business.

"It's a well-known conjecture that has received a lot of attention over the years, and surprisingly the proof (was) short," said Noga Allon, a mathematician at Princeton University.

Alon described this achievement as "another impressive example showing that AI tools will change - and are already changing - mathematics research dramatically."
The hypothesis relates to graphs. A "graph" is simply a set of points (called vertices) connected by lines (called edges). Mathematical proofs related to graphs have been used in many real-world networks, including the internet.

During the 1970s, a number of mathematicians independently postulated that almost any conceivable graph possessed a double cover for cycles.
A cycle is simply a path within a graph that ends at the point where it began, while a cycle double cover refers to a special set of closed paths that cover the entire graph exactly twice, so that each edge is part of only two closed paths.

Over the years, several prominent mathematicians have attempted to prove this idea put forward by their predecessors. They have indeed succeeded in proving it in a number of specific cases, but they have never been able to prove it in its general form.
The proof produced by artificial intelligence seems to have settled the matter, as it shows that any graph to which the intuition applies can be covered twice using no more than eight carefully selected loops.

Andrew Sutherland, a mathematician at MIT, said that once a problem is perceived as difficult, experts and students may reduce the time they allocate to it. This can make the perceived difficulty of the problem a self-fulfilling prophecy.

He adds: "I think we will continue to see examples of this - problems that are supposedly 'difficult' and have 'easy' solutions that have been found by machine learning models."

OpenAI published the text request that led to the successful proof. This request reveals some of the unconventional tricks engineers discovered to get chatbots to handle mathematical problems.

Sutherland said: "Most of that demand was a guiding framework intended to push the big language model to actually make the effort needed to solve the problem."
The engineers directed the model to distribute tasks among up to 64 agents working in parallel. Combining the efforts of multiple AI systems that communicate with each other is a well-established method for reducing false proofs and fabricated references.
They also asked him not to dismiss the issue simply as unresolved, and therefore unlikely to have a solution.

Perhaps most importantly, they gave him clear encouragement, as the request stated: "Spend at least 8 hours on this matter before you even think about going back or giving up."

This is consistent with the personal experiences of many mathematicians when using large linguistic models, where artificial intelligence sometimes tries to evade the search for new proofs by pointing to failed human attempts.

It seems that the correct approach to dealing with it is very similar to the approach of a teacher in school, which is a combination of positive encouragement and firm instructions.


Post a Comment

0 Comments