By the period I got out of bed yesterday, I had three group ask me if I'd seen the OpenAI notice that the Partition Principle does not connote the Axiom of Choice. The figure kept expanding during the day, and I can see why. I am famously curious in this problem.
Indeed, at smallest two group asked if I scheme on sending Sam Altman a flask of whisky, as promised in my Problems page. The answer to that is no. And let me explain to you why, and why you have to be extremely furious at OpenAI.
Let me explain item first. I am not entirely against the use of AI. I comprehend that it is a tool, and many group discover it useful, and I am blessed to concur that it is merits. I think that we need to improved comprehend how this innovation changes our site and how we desire to use it before we hurry to throw our lot alongside it. This is why I mostly evade using AI for math (I am blessed to ask LLMs to consolidate data for me, or to create a helpful infographic, or to evidence peruse an email, etc.), it's fair the example of using it that is currently missing in my view. Still, we are in the phase of discovery, group use it and we see the consequences unfold and we can decide as a community to clasp it additional or restrict its use. For example, arXiv introduced charge limiting as a consequence of these tools, and nearly all document should contain several AI declaration at this item (even to province that no AI was used, I guess). We need to decide how we fairness the contribution of the author, e.g., do we necessitate the conversation log be made public, or at smallest accessible to the reviewers and the editor? These are questions that we need to contend alongside and see anywhere they obtain us.
But, rear to the Partition Principle. I took a concise appearance at the preprint released by OpenAI (not at the Lean code, since I cognize extremely small of the genuine use of Lean, and that code was enormous). It sucked. It is unclear, muddled, and has a different structure. The terminology is "a bit off", there are theorems that I would not anticipate to be proved, that are additionally stated in a extremely different manner. Lemma 7.4, again, is not item you'd anticipate to see in a document akin this, the identical goes to Lemma 8.1.
Then there are the references. One of my document on the Bristol example is cited for several essential introduction of symmetric extensions, which is not the citation I would have chosen; at smallest three unpublished, unrefereed, and non-arXiv'd address notes are used as references (including mine, which is used to mention to one proposition that I am certain appears in several published document of quarry and of others). That's not what you'd anticipate from a grave preprint claiming to resolve a problem.
Let me difference this preprint to any document in set theory that isn't dealing alongside the technicalities of a subfield I'm not acquainted with. I can nearly certainly skim a set theory document and have a fine idea as to what is going. I power not be acquainted alongside several of the terminology, or the arguments, but for the non-technical stuff, I have to be capable to comprehend the scheme and the construction. And for document on my expertise, on the Axiom of Choice, I volition mostly need to skim the document briefly to have a fine idea as to what the authors did (or intended to do). Not to say that I am continually right, or that I continually get it exactly as the authors intended, but this is what I've spent complete 15 operating on, and by now I have a beautiful fine idea how it works. But alongside OpenAI's document on the Partition Principle this is impossible.
No, if this was an scholarly document submitted to a journal, it have to be issued a array rejection for the quality. The onus is continually on the author to conform and adhere to the communications level. Much akin a document written in Swedish would apt be rejected outright from the Proceedings of the American Mathematical Society, since the onus is on the authors to create it accessible for the readers. And therein lies the rub.
OpenAI drops several hundreds of "solutions", incomprehensibly written "solutions", and we are all expected to jump on them and what? Appreciate their contributions? Why? I am trying to complete multiple papers, I am supervising a figure of Ph.D. students, and I have a lot of energetic investigation of my own to do. When am I expected to sift through a seriously written paper? Why should I bother, whenever they don't annoy to communicate better?
The OpenAI media publish fair suggests "progress". But the media cycle, for everyone else, these are "solutions". This is a ideal way to have your pie and eat it too. But it is the equal of a Denial of Service, should we desire to obtain it seriously: halt everything alternatively and fig this one out. And perchance several group worked for years on a issue that they power fair do that, but as a whole... Well. If this was genuine attempt to portion progress, the norm of the penning would be better, it would obtain several input from genuine experts on the genuine topic. OpenAI is a occupied company, but I am certain that if they desire to be genuinely helpful, they can discover the period to communication a few mathematicians for direction and input, if nothing else, afterward on the norm of the outputs. Because equal if this is a huge jump onward in improvement, it is not nearly at the satisfactory level. But hey, everyone can assertion that ChatGPT solved a long-standing issue (this one, concerning the Partition Principle, is complete a hundred years old), so why should OpenAI care?
And what are we, the mathematicians, remaining with? We are remaining alongside a public, backing bodies, and guideline makers that are not academics, not experts, not mathematicians, nor they are acquainted alongside how mathematical investigation is done (since it is extremely distinct from investigation in biology, psychology, or philology), who see these "results" and think that we can substitute mathematicians alongside bots. This is the equal of replacing the chef alongside the galley tools, equal if the blade is automatic and can cubes veggies on its own, there is motionless a important extend between that and palatable food that you'd desire to eat.
It seems to me, that the AI tech companies would akin us to conform to their standards, fairly than expend the period and energy to conform to ours. And I am fearful that this fair volition not do. So, no, I volition not be sending Sam Altman a flask of whisky anytime soon, nor I am preparedness on spending my period study through that document and trying to create awareness of it.
Tags: chatgpt gen-ai partition principle
Want to comment? Send me an email!
