The Yang-Mills Publicity Stunt

The “shepherds” of the OpenAI Yang-Mills proof yesterday released the Lean code of the proof, together with a validation and an assessment.

The OpenAI press release of October 26 claimed to have proved “Existence of quantum chromodynamics and completions of the Standard Model”, but the “shepherds” have not provided anything at all about this, not even an acknowledgement of the existence of such proofs or any plan to release any information about them. As far as I can tell, the claim to prove existence of “completions of the Standard Model” is likely extremely misleading. Perhaps they just have something about the minimal SU(5) GUT, without any Higgs sector. Even with a Higgs sector, such a theory disagrees with experiment, so in no sense is a viable “completion of the Standard Model”.

The press release goes on to claim:

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.

Now that the actual result has been released, it is clear that this claim is also extremely misleading. The Jaffe-Witten writeup of the problem explicitly states (page 12)

We specifically exclude weak-existence (compactness) as the solution to the existence part of the Millennium problem, unless one also uses other techniques to establish properties of the limit (such as the existence of a mass gap and the axioms).

This is exactly the sort of “existence” proof OpenAI has produced, and they have not separately established the mass gap or properties comparable to it.

This whole story is starting to remind me a lot of the Google quantum computer wormhole publicity stunt. We’re seeing the same sort of fraudulent claims now being widely distributed in the media, e.g.:

Surely there’s a lot more of this kind of thing to come.

The field of theoretical physics has suffered a great deal in recent decades from the problem of misleading hype. Getting in bed with the less than honest efforts of an organization trying to justify a valuation at greater than \$1 trillion is a terrible idea. I believe the group of physicists doing this should publicly explain that the OpenAI press release was misleading, release all the documents they are holding onto, and then shut themselves down.

In any case, now that they have released the Lean code, their plan to keep secret the 1147 page document explaining the proof and devote great efforts to understanding it and explicating it themselves makes no sense. Within hours of the code release, other groups (e.g. here and here) started using AI agents to derive similar documents which they are making public.

Maybe someday AI agents will generate actual “blockbuster” results on this or another fundamental physics topic. Until this happens, theorists should be helping to stop OpenAI’s hype campaign, not encourage it. I’m actually on a vacation trip at the moment, and feel that I’ve now done more than my fair share of this. So, for a while, I’ll only have minimal time to devote to dealing with comments, further developments or addressing the details of what’s wrong with this hype campaign.

Update: A statement from a large group of prominent theorists has been released criticizing the practice of not releasing publicly results such as the Yang-Mills one.

Update: In latest Yang-Mills publicity stunt news, OpenAI has started promoting its models with a new science blog, first entry by Nathaniel Craig about his work on chiral gauge theory. The relation to the publicity stunt is described at the end:

The construction also leaves open the question of defining the interacting theory in continuous spacetime. OpenAI’s recent proof of the existence of Yang-Mills theory (of which the strong and weak forces of the Standard Model are examples) provides the missing puzzle piece. This enabled them to prove the existence of the simplest unified Yang-Mills theory that includes the gauge fields and fermions of the Standard Model by starting with overlap fermions on the lattice and taking the continuum limit.

This “simplified unified Yang-Mills theory” seems to be a more honest characterization of what the OpenAI press release deceptively described as a “completion of the standard model”. The “shepherds” are still keeping everything about this result secret. Supposedly it is an “existence” result, but we don’t know what it is that exists and what “existence” means here.

Craig says that this depends on his work, but is careful to describe his results as only “a step towards defining” or “a candidate for” a theory. In addition, he notes that he has not shown reflection positivity for this construction, and there’s a new preprint out arguing that this kind of construction cannot satisfy reflection positivity at finite lattice spacing (I’m confused about what is known about the issue of reflection positivity holding only in the continuum limit).

Reflection positivity is about the existence of an Osterwalder-Schrader reflection operator that is the Euclidean version of complex conjugation, with reflection positivity corresponding to unitarity. It turns out that I’ve been thinking a lot about this reflection operator for chiral fermions, in two dimensions and in four dimensions (using twistor geometry). I’ll write more about this story in detail, hopefully sometime soon.

This entry was posted in AI. Bookmark the permalink.

10 Responses to The Yang-Mills Publicity Stunt

  1. Sheep says:

    At another GitHub link (https://github.com/yang-mills-group/.github), the following is stated: “Our results categories are YM (pure Yang–Mills theory), QCD, and chiralSU5 (chiral SU(5) gauge theory). Each category will bring together the relevant submissions, independent checks, explanations, and assessments.” Since I don’t see any mention of QCD or chiral SU(5) gauge theory in the current submission, I am led to believe that it will be a part of OpenAI’s November 4th submission (https://github.com/yang-mills-group/YM/tree/main#results-and-assessments), which has not been assessed/released yet.

    Disclaimer: This is speculation.

  2. Alex says:

    I have translated the Lean code into a human-readable version, available here: https://alexisdpc.github.io/yang-mills/

  3. @Alex respectfully, it would be more readable if it weren’t in an AI-generated-website style, and just actual prose in the format people are used to reading.

  4. Peter Woit says:

    Alex,
    I just took a quick look, wasn’t bothered by the format as David Roberts was. It’s remarkable that this sort of explanation of what is going on with this proof can be generated in this way.

    I’m starting to believe that, long before the “shepherds” make any progress at deciphering the 1146 pages of slop that they are keeping secret, other projects like this will have produced much more readable and useful accounts of this proof.

  5. Gil Kalai says:

    Hi everybody. Some comments on the new development.

    1) It seems that OpenAI (and possibly Anthropic as well) were able to make substantial progress on the major mathematical physics goal of finding a rigorous mathematical formulation of quantum field theory. This represents a major breakthrough.

    2) It looks overall clear that AI mathematical capabilities in proving theorems and solving problems exceed those of human mathematicians across most mathematical fields. This by itself is perhaps among the most amazing scientific and technological advancements of our lifetime. (While this is a strong statement, I don’t think it is “hype”.)

    Here are some additional points:

    a) To the claim that the solution does not go as far as we could hope: It looks like the solution went very far. (Nonuniqueness and no mass gap notwithstanding.) Of course, it needs to be verified and understood. And we can expect more advances later on.

    b) I agree that the fact that the main 1200-page paper is not publicly available is unfortunate. Since our community is confused regarding the new reality of AI in mathematics, it is natural that some proposed approaches are worse than others.

    c) The ability to juggle between formal (Lean) verification and human-style written manuscripts is surprising and remarkable. This seems evident from Armstrong’s and Hairer’s efforts and seems quite important independently of QFT. (So this is a positive indirect consequence of item b.)

    d) The implications for theoretical physics (if any) are unclear. Some physicists regard rigorous mathematical foundations of QFT as the basis for further progress in theoretical physics, while others are skeptical. I would speculate that the simplicity of the mathematical framework (or parts of it) might be a crucial factor for applications in physics.

    e) The new developments in AI and mathematics pose serious threats to the community of mathematics, to the culture of mathematics, and perhaps even to the mathematical discipline itself (especially in pure mathematics, and especially for young researchers). I am personally optimistic that mathematics will prevail.

    f) Hype and “publicity stunts”: Peter often sees matters through the lens of “hype” and the need to “fight” hype. In the case of AI and mathematics, this point of view may miss the essence of things.

    I expressed similar views regarding earlier AI progress in mathematics here in June: https://www.math.columbia.edu/~woit/wordpress/?p=15705#comment-253820

  6. Jon says:

    I would be interested if anyone commenting positively about, or generating, these “human readable” summaries has actually found them useful for developing real understanding of what goes on in the construction (or if this has been the case for other examples of long/sprawling AI generated proofs).

    The relatively short papers (~30 pages) I have tried to read from the OpenAI release tend to be poorly written, but one can extract something from the “human readable” versions. The constructive QFT (O(3) model) papers, however, serve only to give reference anchors for prompts to your own local AI. It is remarkable that the impenetrable writing is in some way internally consistent, such that your local AI will give (seemingly) sensible answers. I am not sure how well this works if you don’t already have some sense of the proof strategy, e.g., by familiarity with prior constructive QFT literature.

  7. Peter Woit says:

    Gil Kalai,
    In this case I was using “hype” to be kind. What OpenAI and the shepherds have done here is much worse.

    Questions of how important a result are can be a matter of opinion, but either you have a proof of a mathematical theorem or you don’t. In this case, OpenAI has publicly claimed that they have proven the existence part of the Yang-Mills Millennium prize problem statement. This is a lie. The weak existence result they may or may not have actually proved is explicitly ruled out as a solution to the problem in the problem statement. Furhermore, I think it is highly likely that their claim to have proved existence of a QFT that is in some sense a more complete version of the Standard Model is also a lie. Here though, they have not released the actual result they supposedly have a proof for.

    This result is being used to promote the capabilities of their product in the context of an effort to raise money from investors. In this context, lying about this is fraud. By putting their names and reputations behind this, the “shepherds” are making themselves accessories to the fraud.

  8. martibal says:

    A theory with a non-unique vacuum does not satisfy Wightman axioms, or any other axiomatic description of QFT, right ? The uniqueness of a Lorentz invariant vacuum is, somehow, the first condition required for a qft, right ?

  9. lun says:

    Would you call this a publicity stunt (it is a topic you are worried about, and was posted on arxiv)
    https://openai.com/index/step-toward-defining-the-standard-model/?

  10. Peter Woit says:

    lun,
    It’s part of the same publicity stunt. Will add something about it to the posting when I have time, probably tonight.
    Also, this story involves one of my favorite topics, wick rotation of chiral spinor theory.

Leave a Reply

Informed comments relevant to the posting are very welcome and strongly encouraged. Comments that just add noise and/or hostility are not. Off-topic comments better be interesting... In addition, remember that this is not a general physics discussion board, or a place for people to promote their favorite ideas about fundamental physics. Your email address will not be published. Required fields are marked *