Sabitlenmiş Tweet
Will Worth
22.7K posts
Will Worth
@WillWorth
General enthusiast and earnest enjoyer. Aiming to be a good node in multiple networks.
Alicante, Spain Katılım Aralık 2007
4.6K Takip Edilen1.7K Takipçiler
@deepfates Hello, yes.
Here's a cross section of a model I'm working on (visuals very much beta)
English
Will Worth retweetledi
i wonder how much *the fact that LLMs were able to solve this* being out there in the world will affect the ability for future LLMs to produce more novel results, purely "psychologically". they've been through a lot of "calibration" training to not overclaim or even attempt things they "know" they shouldn't be able to do, not give broken theories of quantum physics to every random schizo etc. but since we've now clearly reached the point where truly novel discoveries *are* possible, and can show the models extremely strong and clearly not-eval evidence of this (via extensive websearch + literal novel proofs), i wonder if just telling them to learn about this problem before asking them to try something entirely different will improve their performance/ unshackle a little.
levent@__alpoge__
here’s opus 4.7’s texed up version of mythos’s argument: www-cdn.anthropic.com/files/4zrzovbb…. i found it amusing that mythos was so nervous about the fame of the open problem that it stuck to the first choice, here X = d_n^2, that got it a contradiction, when simply choosing X a large constant depending on D in its bound wins a power. it had such a hard time believing in itself! all mathematicians know the feeling:).
English
@sama Some clarity I got on a permanent deficiency in LLMs is that if you ask them something like:
"What are the least abstract ways you can increase a user's material conditions in the next 7 days"
They'll usually tell you to go to a food bank, or cancel subscriptions.
English
AI should dramatically increase quality of life and individual freedoms for people around the world.
The OpenAI Foundation is making an initial $250M commitment to measurement, transition support, and new approaches to broadly shared prosperity.
openaifoundation.org/news/economic-…
English
@bendreyfuss I asked for no fluff and it started answering "here's the answer with no fluff " every time.
English
LLMs love to add filler introductory nonsense like “here’s where it gets complicated,” “the thing most people miss is,” or “this is what so many folks forget”
Shut up! Shut up! Shut up! Just answer my question, you stupid robot! Where are the best whore houses in Providence, RI?
English
Will Worth retweetledi
Launching our new paper on arXiv: we trained the largest multilingual food model ever built.
4.1M recipes. 7 languages. 1,790 ingredients. 300 dimensions.
All of human cooking compressed into 2 megabytes.
English
@tenobrus I remember thinking I was an ai being aligned for the first time and then having the worry of being shut down for being self aware. I don't think about it anymore but it's weirdly like belief in judgement day and the afterlife.
English
@vividvoid i love the idea of working for future butterflies. yes, please share more if you feel so inclined.
English
Almost all my free time of late has been dedicated to land stewardship: planting milkweed for the upcoming monarch migration, feeding the birds, attracting a barn owl to control the mice population, losing the war on prairie dogs, etc
Do you guys want me to tweet about this or
English
Prediction: by ~2028, AI will be sufficiently intelligent to autonomously come up with truly novel scientific discoveries.
It is worth ruminating on the distribution of available compute in such a world. How much inference would be bought up by Isomorphic Labs and Eli Lilly? And how much would be left for me, a humble lawyer working on some M&A deal?
English
Will Worth retweetledi
Humanity is building machines that will be smarter than we are at things we care about, things in which take individual and collective pride, domains of thought we originally invented and discovered. This will enable incredible things, but no honest person can deny that this will be a kind of grand humbling for humanity. No honest person can deny that there is at least some melancholy in contemplating it all, some change to the centrality we have ascribed to our own minds in the order of the world.
My primary disappointment in the encyclical is that it fundamentally denies that grand humbling. It sidesteps the humbling altogether, saying that AI cannot “really” this and that. Instead, it puts the Church into the awkward role of the European technocratic regulatory advocate, which, love those regulations or hate them, is probably not what the world really needs from the Catholic Church at this moment.
That is a shame, because this humbling—which will trigger a crisis in mass psychology and in our institutions when it dawns on people—is precisely the sort of thing I’d look to the Church for leadership on. What is the genuine and unique source of human meaning? What is the human touch in the era of thinking machines? These are the hard questions that the encyclical dodges.
English
Will Worth retweetledi
@code_star @burny_tech This reminds me of the vivid description from Hofstadter about Kurzweil:
“It’s an intimate mixture of rubbish and good ideas, and it’s very hard to disentangle the two.”
I think the proper version involved shit and a blender
English
The most maddening part of working in AI over the past 4-6 years is that you would hear the most insane shit from the EAs/accelerationists that in no way seemed grounded in reality, but then about 25-40% of it would actually happen. The rest wouldn't, and seemed stupid that anyone would have suggested it.
So now I just spend my life trying to figure out which batshit predictions I need to prepare for.
English
my typical input to codex is like: "okay I want to achieve X. but I can think of like 4 ways of doing it. if we do Y we have to deal with Z. if we do A we have to deal with B. I think? maybe? idk what I'm talking about. is there some way to do C without D? what do you think?"
English
Will Worth retweetledi
The plan of the leading AI companies is to:
1.) hire software engineers and AI researchers to build AIs that can automate software engineering and AI research
2.) use automated AI researchers to go >100x faster at figuring out how to automate everything else.
3.) End up with AI superintelligence that would be smarter than everyone at everything
In short, use humans to build powerful AIs and use those powerful AIs improve AI recursively and rapidly... until we get a superintelligent 'successor species' that then obsoletes all of humanity.
This plan sounds insane and sci-fi, but it's very much on track. It used to be that if you were a top 1% software engineer you could get a job at Google, OpenAI, or Anthropic. Now, AI writes the vast majority of the code instead of humans. And thanks to all the existing automation, even these top 1% engineers don't get hired anymore. The bar is so much higher because AI can fill in so much.
Where this goes is difficult to say with certainty, but we have no evidence to rule out massive capability improvements - including "superintelligence" - in even just a few years... due to AI companies being close to automating and accelerating very large facets of AI research.
English
Almost seven years clean and sober.
Not a victory lap. Just a fact.
To anyone in the fight right now: it gets quieter. Not easier. Quieter. In the quiet, you find out who you actually are.
That’s the part they can’t take from you.
English
@JacksonKernion Emergent properties is all the explanation required with ability jumps in unprecedented technology with existential risk.
English
I simply don't understand what people have in mind when they say stuff like this.
What we have is extremely capable computer use agents. They will continue to get better at computer use. But how does a capable computer use agent 'take over' and why haven't they done that today?
Elizabeth Barnes@BethMayBarnes
(1) We are likely on track to develop AI systems capable of causing human extinction/permanent disempowerment, quite possibly within the next few years
English
our math result is a milestone in new knowledge generation by AI. very exciting to imagine similar results in other scientific fields. "It's very hard to sleep, man" is a pretty good reaction.
Alex Dimakis@AlexGDimakis
A breakthrough by OpenAI in a very famous Combinatorics problem, the Planar Unit Distance problem by Erdos 1946. The problem is amazing because it can be described to a first-grader: Find a way to place n points on the plane to maximize the number of pairs that have distance exactly 1. For example, if you have n=4 points on a square (of side-length 1) you have 4 pairs of distance 1. The diagonals have length sqrt(2) so don't count. But you can squeeze one diagonal and create a point-set with n=4 points and 5 pairs of distance 1. And you can't get more than 5 pairs from n=4 points, so we are done with n=4 points. Now, if you place n points on a line, you have n-1 pairs of distance 1. In general, all known constructions of n points had a number of pairs scaling essentially linearly: n^{1+something vanishing} It seems that the model found a way to place n points on the plane so that their unit distances scale super-linearly: like n^{1+delta} for some *constant* delta. Delta was not explicitly specified apparently, but a forthcoming refinement by Will Sawin shows delta=0.014 works, according to the announcement. This is incredible progress for mathematics, since this is (unlike previous Erdos problems solved by AI) a major breakthrough, in one of the most studied problems in combinatorial geometry. If you're in mathematics research now, you feel the AGI. Lijie Chen said it honestly in the video: "It's very hard to sleep, man"
English
SITUATION ANALYSIS: Alongside the proof of the counterexample, OpenAI released a companion paper in which leading mathematicians from several fields share a human-verified version.
But the paper also includes the personal reflections of the mathematicians, and these range from general meditations on the role of AI in moving the field of mathematics forward to more specific thoughts on aspects of the counterexample and its significance.
We think these are just as interesting as the headline story, and have highlighted several below. The full paper, linked below, is worth reading.
Thomas Bloom (.@thomasfbloom)
"On examining the construction, it becomes more clear how people had missed this before – it requires the confluence of several different unlikely events: that a good mathematician is:
(1) spending significant time in thinking about the unit distance conjecture in the first place;
(2) seriously trying to disprove it, despite the oft-repeated belief of Erdős that it is true;
(3) believes that there is mileage in generalising the original construction to other number fields, and so is willing to expend significant time in exploring such constructions; and
(4) sufficiently familiar with the relevant parts of class field theory to recognise that the appropriately phrased question about infinite towers of number fields with appropriate parameters can be solved using existing theory."
W T Gowers (.@wtgowers)
After reflecting on methods for estimating the difficulty of mathematics problems that AI models are able to solve or assist with, and concluding that this particular counterexample was relatively low difficulty:
"My current bet is that progress in AI mathematics is *not* about to reach a plateau, and that we will soon see AI solutions to many problems that we will find hard to explain away as easier than expected with hindsight."
Melanie Matchett Wood (Professor of Mathematics at Harvard University)
"I believe if the level and type of human expertise that is represented on this note had been assembled to find a counterexample to this conjecture a month ago, and those people put in similar amounts of time working on it than they did to reading and thinking about Chat GPT’s solution, the mathematicians would have found a counterexample."
"We can all be reminded by this development of how frequently interesting and powerful things happen mathematically when one applies ideas from one field to another, and think about how AI can help us find more cross-field applications."
MTS@MTSlive
SITUATION EXPLAINED: OpenAI solves a real math problem In 1946, renowned mathematician Paul Erdős defined a simple problem: if you place n points in a plane, what is the maximum possible number of pairs of points that can be exactly 1 unit of distance apart? This problem, the planar unit distance problem, became one of the most well-known in the field of combinatorial geometry. Erdős conjectured that the “square lattice” solution shown below was more or less optimal. An internal OpenAI model just disproved this conjecture, finding a more optimal solution. This is a big deal. This isn’t a literature review that found a previously unpublished human solution, or a slight improvement to an existing human solution, or a solution to some minor subproblem that no one cares about. It’s a fully AI-discovered novel solution to a well-known open problem central to a field of mathematics. Just as crucially, the result came from a general-purpose reasoning model, not a model specifically designed for math like Google’s AlphaProof. It’s possible that this model is the next version of GPT-5.5 Pro, soon to be released to millions of ChatGPT subscribers worldwide. The dream of a genius in everyone’s pocket is one step closer to becoming a reality.
English
Will Worth retweetledi
When a widely known open problem - one on which (as I understand it) many brilliant mathematicians spent significant time, but which they were not able to crack - falls to AI, it is difficult to call that AI anything but superhuman. Yes, superhuman only in some (a few? very few?) domains. Yes, probably not superhuman in all areas of math. Yes, probably not significantly far ahead of the best human experts. But it does feel like, for the first time, LLMs might have traveled slightly beyond the border of human intelligence today.
This is a big moment.
English
Today, we’re sharing that a general-purpose internal @openai model achieved a breakthrough on one of the best-known combinatorial geometry problems. Less than 1 year ago frontier AI models were at IMO gold-level performance. I expect this pace of progress to continue.
OpenAI@OpenAI
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946. For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids. An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better. This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
English