Latest Notable AI accomplishments

  • Topic: AI 
  • Thread starter Thread starter gleem
  • Start date Start date
  • Tags Tags
    Ai
Join the discussion
Registration is free. Ask a follow-up in this thread, or start your own.
Sign up Log in
137 replies · 35K views · last reply
Physics news on Phys.org
722 new papers by OpenAI
Full list
The vast majority of results were obtained with the same procedure using an unreleased internal OpenAI model. On average, each result used three hours of ChatGPT Pro thinking compute with that model. Over the course of the evaluation, the model was posed approximately 4,000 problems. Aggregating the output into result families and manuscripts and requiring an appropriate level of significance led to the catalog outlined above.
 
Reply
  • Like
  • Informative
Likes   Reactions: Demystifier, Frimus, Greg Bernhardt and 1 other person
The next question is how these solutions are to be celebrated, if at all. OpenAI is not a math research foundation. While they may have mathematicians on staff, they have not released their solutions in a currently accepted, peer-reviewed form. Questions about unexplained, unusual methods and plagiarism persist.

Mathematicians are becoming increasingly unhappy with these bulk releases of math solutions.
https://www.wired.com/story/openai-is-pissing-off-a-bunch-of-mathematicians-again/
“Math by tweet and math by press release to me is not the way to nurture the ecosystem that created the fertile ground that they have trained on,” Northwestern University mathematician Bryna Kra

Supposedly, some employees at OpenAI have declared that math is dead. How many grad students' careers have been upended by the corporate exploitation of math problem solutions with their AI models? How many UGs are reevaluating their career? But then again, who needs these "would be" mathematicians? :wink:
 
gleem said:
Questions about unexplained, unusual methods and plagiarism persist.
They persist for the Navier-Stokes solution (and many other suspicious incidents), but not for these 722 new papers. Also, 3 hours of a single agent per result is much more acceptable than 88 hours of 10000 agents for one brute-forced result.

Mathematicians have every right to be pissed-off about the unfair $15 million dollar scoop. But they also accept that OpenAI tried to behave fair this time, and did coordinate with the „representatives“ the mathematicians had quickly assembled.
 
Last edited:
gleem said:
How many grad students' careers have been upended by the corporate exploitation of math problem solutions with their AI models? How many UGs are reevaluating their career?
Since Anthropic is next to go IPO (I guess mid November 2026), it is understandable that OpenAI tries really hard to showcase the strength of its internal models. If Anthropic manages to push-through its targeted $2 trillion market capitalization, OpenAI‘s IPO in 2027 could become a disappointment.

Maybe OpenAI targeted the „low hanging fruits“ for its current internal model, to make it at least a bit more difficult for Anthropic. I am curious what we will see from Anthropic before their IPO.
 
gentzen said:
Also, 3 hours of a single agent per result is much more acceptable than 88 hours of 10000 agents for one brute-forced result.
What's the difference?
 
gmax137 said:
What's the difference?
3 hours of a single agent is affordable, while 88 hours of 10000 agents burning $15 million dollar is not.
https://agmai.org/general-sep29/

3. Ensuring broad access​

[…]
As for publicly available models, unequal access due to economic and other factors risks establishing a multi-tier hierarchy across the mathematical world and greatly exacerbating existing inequalities that arise from institutional wealth and technological privilege.
 
From my understanding, the Riemann Hypothesis critical strip paper would be huge if true. I'm curious how long that one in particular took.
 
An amateur astronmer used Claude and Codex to find an exoplanet candidate. His results were good enough that he was given a slot on the TESS observatory.

 
Borg said:
An amateur astronmer used Claude and Codex to find an exoplanet candidate. His results were good enough that he was given a slot on the TESS observatory.

Using AI as a tool is the way to get ahead, at least for now. The next step in its use is that of an associate. After that, a competitor.

gentzen said:
Mathematicians have every right to be pissed-off about the unfair $15 million dollar scoop. But they also accept that OpenAI tried to behave fair this time, and did coordinate with the „representatives“ the mathematicians had quickly assembled.


It was reported that OpenAI would release its math solutions piecemeal. Dumping hundreds at once on GitHub does not seem well coordinated. I guess one could say that because of the need to verify the results, mathematicians will be busy for a while.

Some believe that AI generated math will require more mathematicians.
https://terrytao.wordpress.com/2026/09/24/were-gonna-need-a-lot-more-mathematicians/
But this author hopes that we will always take AI generated tasks with a grain of salt and verify the validity of its results. The problem is that we are not doing that now to the degree we should, especially in understanding AI itself.
 
AI is significantly improving weather predictions.

OpenAI has developed a model that can forecast cyclone development one more day sooner than before.

https://www.wired.com/story/deepmind-ai-model-can-predict-hurricanes-earlier/
In a paper published on Thursday in Nature, researchers show that the WeatherNext AI model can predict cyclones with unprecedented accuracy. On average, it gives forecasters a day more lead time than existing models; this means its predictions three days out are as accurate as previous models’ predictions two days out. On the ground, that extra day can mean a lot.

A collaboration at MIT has developed an AI system to increase the accuracy of severe weather two to six weeks in advance.

https://news.mit.edu/2026/researche...out-1014?utm_source=substack&utm_medium=email
“Not only does this do better on time averaged forecasts for this range, which is already very useful for farmers and members of the energy industry trying to match supply with demand, but it also improves predictions of extreme weather events, whether it’s flooding or cold spells or something else,” co-author and MIT Research Scientist Judah Cohen says. “It’s really important to predict both, but extreme events might be the most impactful part of this for society.”
 
Some very big results are being claimed in harmonic analysis, computational complexity, and mathematical physics.

Local smoothing and the Kakeya conjecture ...​


Our models produced a proof of the local smoothing conjecture in every dimension and therefore also resolve the Kakeya conjecture and Stein’s restriction conjecture in all dimensions. The new proof is based on a novel auxiliary partial differential equation

The Exponential Time Hypothesis is false...​


The ETH has been widely believed to be true, and has been used in hundreds of papers to derive conditional computational hardness results across a wide variety of fields including graph theory, computational biology, databases, computational geometry, and more. Surprisingly, our system has recently proved that the ETH is false, a result that was subsequently verified in Lean.

Existence of quantum chromodynamics and completions of the Standard Model ...​

The physics collaboration will understand and explain a result that gives a mathematically rigorous existence proof of certain interacting quantum field theories containing gauge fields and fermions in four dimensions, including Quantum Chromodynamics (QCD) and a unified completion of the Standard Model of particle physics. Our work builds on Balaban’s approach⁠(opens in a new window) in constructive quantum field theory. As a subproblem, this work also includes a proof of the existence of pure Yang–Mills theory, which is part of the Quantum Yang–Mills Theory Clay Mathematics Institute Millennium Prize problem⁠(opens in a new window).

https://openai.com/index/building-scientific-understanding-with-academic-communities/
 
Matterwave said:
The Yang Mills one was a surprise to me. But also, I had no idea Balaban had some giant tome from the 80's dedicated to the problem lol. I'm eagerly awaiting David Gross' groups analysis of the output.
I think we have to be extra skeptic with Yang-Mills solution even if Gross et al. give it validation. The Navier-Stokes (NS) problem was a straighforward (but hard) problem of differential equations and even then we are not sure if it is solved or not. It will take time to decipher what the swarm of bots did. The Yang-Mills problem is a different beast altogether because you have to show that a construction of QFT works before even attacking the problem. Whatever the solution, it will take more time to analyze than NS.
 
Last edited:
pines-demon said:
I think we have to be extra skeptics with Yang-Mills even if Gross et al. give it validation. The Navier-Stokes (NS) problem was a straighforward (but hard) problem of differential equations and even then we are not sure if it is solved or not, it will take time to decipher what the swarm of bots did. Yang-Mills is a different beast altogether because you have to basically show that a construction of QFT works before even attacking the problem. Whatever solution, it will take more time to analyze than NS.
Yes, all the AI results need to be checked carefully and rigorously. And just to be clear, the Yang Mills one is a partial result at best. The claim is only "a mathematically rigorous existence proof of certain interacting quantum field theories". There's no claim of hitting all simple compact gauge groups or proof of mass gap existence as the problem requires.

Still, it's very exciting if we finally have one rigorously proven interacting QFT haha.

But I am curious about your "even if Gross et al. give it validation" point. That group seems very legit and relevant since Gross was one of the founders of QCD. Are you saying that as a general word of caution or do you worry about that group in particular or their incentives? Maybe Gross will be biased because he seriously wants a rigorous foundation for QFT?
 
Matterwave said:
But I am curious about your "even if Gross et al. give it validation" point. That group seems very legit and relevant since Gross was one of the founders of QCD. Are you saying that as a general word of caution or do you worry about that group in particular or their incentives? Maybe Gross will be biased because he seriously wants a rigorous foundation for QFT?
Maybe I am overworrying about his age, Nobelitis and his relation to OpenAI. At least they announced other big heads in there (although Jaffe is also as old). But even if they all agree with it other physicists will maybe not agree with their construction of QFT. Why not bring more people? Why not bring Tadeusz Balaban (whose work was used to contruct all of this) into this task for example?

Another point that worries me is that the proof is not public yet. It think they can still analyze it even if they publicly release it.