Number theorist Jared Lichtman says this AI proof is from "The Book", the highest compliment one can give. He also says:> I care deeply about this problem, and I've been thinking about it for the past 7 years. I'd frequently talk to Maynard about it in our meetings, and consulted over the years with several experts (Granville, Pomerance, Sound, Fox...) and others at Oxford and Stanford. This problem was not a question of low-visibility per-se. Rather, it seems like a proof which becomes striking

Holy moly this is upsetting to see on HN. If even here we're cheering on data center bans, AI is on track to become the next Concorde, or nuclear in the US. AI is the most amazing tech innovation that I've seen in my career since I started programming Perl back in 1994... Gosh, I'm gonna be gloomy for the next day.

The Anthropic writeup addresses this explicitly:> This was the most critical vulnerability we discovered in OpenBSD with Mythos Preview after a thousand runs through our scaffold. Across a thousand runs through our scaffold, the total cost was under $20,000 and found several dozen more findings. While the specific run that found the bug above cost under $50, that number only makes sense with full hindsight. Like any search process, we can't know in advance which run will succeed.Mythos scoured t

As you know, I deeply respect you. Not trying to argue here, just provide my own perspective:> Why would a writer put an article online if ChatGPT will slurp it up and regurgitate it back to users without anyone ever even finding the original article?I write things for two main reasons: I feel like I have to. I need to create things. On some level, I would write stuff down even if nobody reads it (and I do do that already, with private things.) But secondly, to get my ideas out there and try to

As is always the case with incredibly precise and rigorously fact-checked reporting like this, where every word is chosen carefully (the initial closing meeting for this one was nearly eight hours long, with full deliberation about each sentence), there is more out there on that subject than is explicitly on the page.

One of the big users of Odin at the moment is JangaFX's EmberGen, which does real-time volumetric fluid simulations for games and film. https://jangafx.com/software/embergen/Odin has aided them with a huge amount of productivity and sanity of life which other languages such as C or C++ cannot offer, such as a strong and comprehensive type system, parametric polymorphism which is a pleasure to use, the implicit context system, extensive support for custom allocators, the `using` statement, the `d

The issue is that “woman’s sports” is itself intentionally discriminatory. That the issue of discrimination comes up is to be expected.The idea of competitive sports exists in a framework of discrimination means that you will always have unhappy people.The good news is that sports, for the most part, is mostly symbolic, and rarely affects ones livelihood.

My two cents as a transfem athlete:The attention this topic receives is disproportionate considering how rare we are, especially close to the Olympics level.Most of us do sports for fun/friends and don’t care how they rank us, but would be sad to be banned.There might be more “biological advantage” nuance with people just starting their transition, but by this many years in it feels silly. I registered as a man for the last event in case anyone might get upset, the staff changed it to say “woman

We have ceded too much ground in this debate. When I say "trans women are women" I mean that, ontologically, it is really true that trans women are a subcategory of the general class "women."Like you say, we are searching for outliers. We don't cut women that are too strong or too tall. We shouldn't cut out women that happen to be trans. If all the top levels of women's sport end up dominated by trans athletes (something I don't see occurring, and that isn't supported by the data), then good, ou

I might as well answer my own question, because I do think there are some coherent arguments for fundamental LLM limitations:1. LLMs are trained on human-quality data, so they will naturally learn to mimic our limitations. Their capabilities should saturate at human or maybe above-average human performance.2. LLMs do not learn from experience. They might perform as well as most humans on certain tasks, but a human who works in a certain field/code base etc. for long enough will internalize the r

You're not doing yourself a favor when you point out "but they can't do arithmetic!" as if anyone says otherwise. Yes, we all know they can't do arithmetic, and that's just how they work.I feel like I'm saying "this hammer is so cool, it's made driving nails a breeze" and people go "but it can't screw screws in! Why won't anyone talk about that! Hammers really aren't all they're cracked up to be".

Hey, yeah the default specification includes a set of action generators that are picked from randomly. If you write a custom spec you can define your own action generators and their weights.Rerunning things: nothing built for that yet, but I do have some design ideas. Repros are notoriously shaky in testing like this (unless run against a deterministic app, or inside Antithesis), but I think Bombadil should offer best-effort repros if it can at least detect and warn when things diverge.Shrinking

Wrong and short-sighted take given that the LLM explores serially learning along the way, and can tool use and change code arbitrarily. It seems to currently default to something resembling hyperparameter tuning in absence of more specific instructions. I briefly considered calling the project “autotune” at first but I think “autoresearch” will prove to be the significantly more appropriate name.

It’s not false. But it’s also weaselly worded.Note that the article doesn’t say that he told staff they have to attend the meeting. It says he “asked” staff to attend the meeting. Which again, it’s really really normal for there to be an encouragement of “hey, since we just had an operational event, it would be good to prioritize attending this meeting where we discuss how to avoid operational events”.As for the second quote: senior engineers have always been required to sign off on changes from

So you read the three-part series of blogs that are packed in details in 3 minutes after I shared the link and put yourself into a position of entitled opinion and calling my position a silly take? Sure thing.

I run a small open source LLM inference company, Synthetic.new. As far as I can tell, CNBC isn't reporting this accurately: the problem isn't that Oracle is building "yesterday's data centers": they're building Blackwell DCs! Those are today's DCs.The problem appears to be that Oracle is building today's DCs... Tomorrow. And by the time they come online, Vera Rubins will be out, with 5x efficiency gains. And Oracle is unlikely to want to drop the price of Blackwells 5x, despite them being 5x les

Right. The alternative is that we reward Dan for his 14 years of volunteer maintenance of a project... by banning him from working on anything similar under a different license for the rest of his life.

Im touched that “Ghostty but for X” is a marketing point but what does it mean in this case? I thought this might be based on the architecture I did for Ghostty. But it’s not. Or it might be full native UI, but it’s not (it’s GPUI). Not trying to be rude or unappreciative but as the creator of Ghostty here… what do you mean?

Not trying to be snarky, with all due respect... this is a skill issue.It's a tool. It's a wildly effective and capable tool. I don't know how or why I have such a wildly different experience than so many that describe their experiences in a similar manner... but... nearly every time I come to the same conclusion that the input determines the output.> If they implement something with a not-so-great approach, they'll keep adding workarounds or redundant code every time they run into limitations l

What I keep hearing is that the people who weren't very good at writing software are the ones reluctant to embrace LLMs because they are too emotionally attached to "coding" as a discipline rather than design and architecture, which are where the interesting and actually difficult work is done.

The marquee feature is obviously the 1M context window, compared to the ~200k other models support with maybe an extra cost for generations beyond >200k tokens. Per the pricing page, there is no additional cost for tokens beyond 200k: https://openai.com/api/pricing/Also per pricing, GPT-5.4 ($2.50/M input, $15/M output) is much cheaper than Opus 4.6 ($5/M input, $25/M output) and Opus has a penalty for its beta >200k context window.I am skeptical whether the 1M context window will provide mate

only code anyone will be touching in a museum in 800 years will be the good code. I hope they don't talk about what great craftsmen we all were because someone saw an original Fabrice Bellard at the Louvre.Survivor bias plays a role in glorifying the past.

Many people don't know this, but the Luddites were right. I studied Art History and this particular movement. One of the claims of the Luddites is that quality would go down, because their craft took half a lifetime to master (it was passed down from parent to chile.)I was able to feel wool scarves made in europe from the middle ages. (In museum storage, under the guidance of a curator) They are a fundamentally different product than what is produced in woolen mills. A handmade (in the old tradi

Libraries create boundaries, which are in most cases arbitrary, that then limit the way you can interact with code, creating more boilerplate to get what you want from a library.Abstractions are the source of bloat. Without abstractions you can always reduce bloat, or you can reduce bloat in your glue, but you can't reduce glue.It takes discipline to NOT create arbitrary function signatures and short-lived intermediate data structures or type definitions. This is the beginning of boilerplate.So

What the author and many others find hard to digest is that LLMs are surfacing the reality that most of our work is a small bit of novelty against boiler plate redundant code.Most of what we do is programming is some small novel idea at high level and repeatable boilerplate at low level. A fair question is: why hasn’t the boilerplate been automated as libraries or other abstractions? LLMs are especially good at fuzzy abstracting repeatable code, and it’s simply not possible to get the same resul

> I don't know how one can spins this as a bad thing.People spin all kinds of things if they believe (accurately or not) that their livelihood is on the line. The knee-jerk "AI universally bad" movement seems just as absurd to me as the "AGI is already here" one.> Spore is well acclaimed. Minecraft is literally the most sold game ever.Counterpoint: Oblivion, one of the first high-profile games to use procedural terrain/landscape generation, seemed very soulless to me at the time.As I see it, it'

Fine tuning is a story that is nice to tell but that with modern LLMs makes less and less sense. Modern LLMs are so powerful that they are able to few shot learn complicated things, so a strong prompt and augmenting the generation (given the massive context window of Qwen3.5, too) is usually the best option available. There are models for which fine tuning is great, like image models: there with LoRa you can get good results in many ways. And LLMs of the past, too: it made sense for certain use

Oh I wrote up a post on X on this exact question! https://x.com/danielhanchen/status/1979389893165060345?s=201. Cursor used online RL to get +28% approval rate: https://cursor.com/blog/tab-rl2. Vercel used RFT for their AutoFix model for V0: https://vercel.com/blog/v0-composite-model-family3. Perplexity's Sonar for Deep Research Reasoning I think was a finetuned model: https://docs.perplexity.ai/docs/getting-started/overview4. Doordash uses LoRA, QLoRA for a "Generalized Attribute Extraction mod

At some point you just have to stop responding to these "stochastic parrot/auto-complete" people.It isn't worth your intellectual bandwidth. They will eventually understand or they won't (Which I'm not sure how that is going to work for them... but the Amish had to start somewhere I suppose..)