p(Dulany) @dulanyw.bsky.social · 38mThe man still hasn't answered how many local models he runs, sus imo 160
Reposted by p(Dulany)⟨ Jargonaut NPC Alex ⟩ @lastnpcalex.agency · 3hwhy don't the big labs treat this down time like telescope time? you make a proposal, they can get a few of their experts to look it over, then burn it out on an allotment of token slack. argument against: it would slow them down, a little counter: slow is smooth, smooth is fast 3193
p(Dulany) @dulanyw.bsky.social · 1h(It feels like "we have some idle cycles, YOLO" to me, but 🤷♂️) What do you think the right way to try/release results would be, in an ideal world? 110
p(Dulany) @dulanyw.bsky.social · 1h(Sorry, 1 more Q, if you're open to it) Imagine tabula rasa - rewind clock before any of OpenAI's math releases They have a model that can solve Millenium problems, but haven't used it yet. Do you think they should try or not? If they try, what would be the right approach to research and release? 110
p(Dulany) @dulanyw.bsky.social · 2hWhere do you think the mathematicians get it right and get it wrong in this situation? 120
p(Dulany) @dulanyw.bsky.social · 4hI don't think the value is binary here - like a great book, there is a lot of knowledge/understanding to be gained from a range of analyses and perspectives on the artifacts. Just because OAI gets some credit, doesn't mean there isn't a lot more to be found there! 130
p(Dulany) @dulanyw.bsky.social · 4hI mean - yes, yes, and kinda (they get credit for expanding the frontier of knowledge, but I think everyone agrees that more work remains to be done to make it universally accessible). With the Lean traces, it's already transmissible between machines. 140
p(Dulany) @dulanyw.bsky.social · 4hI would frame it more as "unfinished product" (valuable, semi-structured, incomplete) rather than "pollution" (useless, amorphous, destructive), but a lot of that depends on perspective, for sure. I'm interested to see what comes of it! 130
p(Dulany) @dulanyw.bsky.social · 4hI dunno, man - I think it's complicated! Review articles are common in academia (and highly cited!), where authors gather and contextualize the existing work. If "understanding" is the key contribution (as many are saying), then the interpretation work is an opportunity to spin straw into gold. 140
p(Dulany) @dulanyw.bsky.social · 5hAnyway, check out their website and FAQ - their lines on AI use are interesting www.ahmath.orgahmath.orgAHM 031
p(Dulany) @dulanyw.bsky.social · 5hThank you! I agree somewhat - I don't know what they're thinking, releasing "results" without Lean proofs. Maybe they believe it provides incremental progress that others can build on? But the ones with proofs - they at least provide a verified path to knowledge, and can be mined for insight. 161
p(Dulany) @dulanyw.bsky.social · 6h"unrefined industrial waste" implies pollution, damage, etc. Do you think OpenAI's behavior is doing actual damage to human knowledge/understanding of math? (not trying to be standoffish - genuinely curious about your thoughts) 3100
p(Dulany) @dulanyw.bsky.social · 7hI wonder at what point the mathematics community overplays its hand here. A lot of universities are facing budget constraints, and when administrators ask "how is your research contributing to your field of knowledge", they won't love to hear "gatekeeping while avoiding technology and discoveries" 1131
Reposted by p(Dulany)Zoomer Antimillenarian @surcomplicated.bsky.social · 7hYou don’t need mathematicians’ permission to do math. The whole fucking point of math is that either the math checks out or it doesn’t. FWIW I’m happy OpenAI released what they have—we should regard it with skepticism until verified more thoroughly but these results are quite useful! 89411
p(Dulany) @dulanyw.bsky.social · 7hUse complex numbers or quaternions, force the players to navigate through the solution space 050
p(Dulany) @dulanyw.bsky.social · 11hRight? "Informally" is pretty load-bearing there, especially in a paper whose sole focus is "accuracy". 010
p(Dulany) @dulanyw.bsky.social · 11hI would absolutely buy that translation from natural-language to Lean would be a lossy process today. It also feels like a very solvable problem with a constrained language set. The more useful path for understanding is Lean->NL, and I'd be shocked if the labs aren't working on that now. 010
p(Dulany) @dulanyw.bsky.social · 07/10/2026Definitely a possible future for more mathematicians, although I don't know about lowered effort/dedication - their work will specifically be bringing the human interpretation/understanding/integration to others bsky.app/profile/dula... 110
p(Dulany) @dulanyw.bsky.social · 07/10/2026The volume of papers is itself a benefit - it gives a wide attack surface to work on If everyone in the field is digging into these 12 results there are another 200 no one is picking up that may have something novel As mathematician Kevin Buzzard said - math is infinite!xenaproject.wordpress.comTo grieve, or not to grieve?Recent events in the field of AI for mathematics have shown us beyond all reasonable doubt that the field is currently undergoing a rapid transformation, unlike anything that we have ever seen befo… 020
p(Dulany) @dulanyw.bsky.social · 07/10/2026(not calling out OP here) but seeing this sentiment a lot today. Mathematics research is turning into physics research - you use a powerful instrument to generate a lot of data, then pan for gold in the results. Academics will do it because that's where the new discoveries will be going forward. 170
p(Dulany) @dulanyw.bsky.social · 07/10/2026I don't, but academics frequently write review articles, surveying existing knowledge in the space without contributing new knowledge. They're great citation bait! Going through the lean code of AI-generated proofs, searching for new knowledge, will be a key way to bring new insight to the field. 000
p(Dulany) @dulanyw.bsky.social · 07/10/2026Why do you think academic mathematicians wouldn't want to do this? 100
p(Dulany) @dulanyw.bsky.social · 07/10/2026There are definitely arguments for more mathematicians bsky.app/profile/dula... 000
p(Dulany) @dulanyw.bsky.social · 07/10/2026Bud, this is my mentions, not a debate club. You've been crashing out all night, I've been responding politely and in good faith, but I'm stopping this here. I'm going to mute you and go sleep in a big bed with my wife. I suggest you go touch grass and/or get the help you need. Godspeed 🫡 000
p(Dulany) @dulanyw.bsky.social · 07/10/2026OpenAI is being deliberately careful about what they say here, because last time they shared info like this, high-profile mathematicians got angry and asked OpenAI to handle the release differently. OAI is now trying the quiet approach vs marketing splash. mathandai.orgmathandai.orgDeclaration — Math and AIRead the declaration and add your name. 000
p(Dulany) @dulanyw.bsky.social · 07/10/2026It only takes a couple of clicks under the "lean" folder, but here, I did it for you: github.com/openai/math/...github.commath/lean/ComparatorChallenges at main · openai/mathContribute to openai/math development by creating an account on GitHub. 100
p(Dulany) @dulanyw.bsky.social · 07/10/2026The reason they are posting these is because they believe these results cannot be found in the literature (novel) and compile to valid results in Lean (verifiable). If you don't like these, you can look at the numerous solved Erdos problems, NS, recent solved ciphers and more. They are easy to find. 100
p(Dulany) @dulanyw.bsky.social · 07/10/2026> "What novel, verifiable pieces of new knowledge have come out of AI in the last month?" > Posts a link to hundred of novel, verifiable new discoveries made by AI in the last month > "That doesn't answer his question" > 🤷♂️🫡🖖 200
p(Dulany) @dulanyw.bsky.social · 07/10/2026This dropped two hours ago github.com/openai/math/github.comGitHub - openai/mathContribute to openai/math development by creating an account on GitHub. 210
p(Dulany) @dulanyw.bsky.social · 07/10/2026You realize this is a very funny thing to post on a thread where you specifically are complaining about accuracy, right 150
p(Dulany) @dulanyw.bsky.social · 07/10/2026It's interesting that you have a bar of "some things beyond a human's capacity" and not "other things that are beyond human capacity" for some reason. It's already impressive that it's doing things beyond human capacity! 120
p(Dulany) @dulanyw.bsky.social · 07/10/2026"Superintelligent" as in "beyond human intelligence". As with many things, Trump claimed credit for things invented by others! 000
p(Dulany) @dulanyw.bsky.social · 07/10/2026Thank you for the invitation - I'll keep that in mind! 😀 The intent of my post was mainly "things that it is possible to know, but humans hadn't discovered yet", like novel answers to math problems or historical codes. It's just starting to expand the knowledge frontier - I'm excited by what's next 020
p(Dulany) @dulanyw.bsky.social · 07/10/2026If you read my post more closely, my primary focus is knowledge humans *don't* know. AI systems are generating novel, verifiable answers to open questions virtually daily now - it's helping us expand the world of knowledge beyond what humans could achieve alone. 420
p(Dulany) @dulanyw.bsky.social · 07/10/2026Fun fact: I'm literally married to a librarian 😂 She may not know everything, but she's always right! 020
p(Dulany) @dulanyw.bsky.social · 07/10/2026A whole library! None have counterexamples to Navier Stokes! 090
p(Dulany) @dulanyw.bsky.social · 06/10/2026One angle: smart folks want to *know*. About. Everything. There are so many questions to ask, so much to understand, that's just beyond human knowledge/capability today Superintelligent AI gives someone to ask, a tool to explore, a way to fill in those knowledge gaps taunting us, just out of reach. 8982
p(Dulany) @dulanyw.bsky.social · 06/10/2026ChatGPT also came up with Florrie "Warrior" and "What Matters" by Laura Mvula, but those don't feel like hard results to me. Good luck in your search - please share if you find it! 010
Reposted by p(Dulany)Fronkon Games @fronkongames.bsky.social · 06/10/2026Capcom announces plans to transform the RE Engine into an AI-Generation Game Engine: ‘Our goal is a future where we create games together with AI’ #gamedev #gamedevelopment #indiedev #capcom #artificialintelligence #aiign.com‘A Future Where We Create Games Together With AI’ — Capcom Announces Plans to Transform the RE Engine Into an AI-Generation Game EngineResident Evil and Street Fighter developer Capcom revealed some big changes coming to its RE Engine, which they plan to gradually improve and transform into an "AI-generation game engine." 0115
