Where the AI market is talking

If math is more than proof, we need to better celebrate the rest of it

I feel that you can see quite strongly the truth in “writing code was never the point” when you encounter inevitably at every company the guy who has been around forever but doesn’t seem to be working particularly hard. Their value is (was) no longer in writing code at a furious pace all day. It was having a coherent, intelligible and communicable theory of the software system the company is founded on. I propose this thought experiment: put all living mathematicians in a very long bus. This bus crashes and they all tragically lose their lives. Can we really say mathematics simply marches onwards with AI alone? Let’s say Anthropic needs a new research result to improve Claude. Are we really already at the point where we burn tokens ad infinitum and arrive at the end of scientific progress…

How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip

I was sincere, because as you point out, the phrasing is not actually ambiguous here. There is only one way to interpret this that is coherent and sensible. The usual % shenanigans weren't even on the table for me. Not that I'd have ever seen "reduced by 0.5x" or any other value below 1x, probably for this very reason. What I do see is "reduced to 0.5x", in which case you're supposed to swap the division for multiplication. Percentages on the other hand are a whole another can of worms, even if these forms are principally interchangeable, and I find them a lot more confusing a lot more often. Not that this would explain the whole mean/geomean thing.

I built non-autoregressive decision models with RL a year ago

I played around with Jev last night and did it for classification tasks that I used Gemini 2.5 flash lite with. It’s a bit faster and bit cheaper, but this is compared to LLM. The consistency was nice to see, BUT, as someone who trained NLP models prior to LLMs, it’s just BERT with more data. I can see why people would want ready made one shot classifier, and I can see the value of sending multiple classifier in one call, but I wouldn’t call it breakthrough. And I believe many labs will replicate it in no time and might have it as part of their harness. I see it as a wake up call for the tech community to go back to basics for most tasks instead of relying solely on generic LLMs.

AI-generated posters don’t have to be horrible

Most AI-generated art is "good enough" for most one-shot designs by now (though still not [always] as good as a good meat-based artist of course) The problems right now are: - Maintaining a consistent and coherent style/aesthetic when you have to generate lots of assets for something. For example sprites in a video game. - Iterating over a design for more than a few passes. For some reason most models keep degrading the entire image if you keep asking for edits to a specific part of the image. Or, some problems are or become impossible for them to fix no matter how you prompt them.

Base Labs launches an open-weight AI safety partnership with Hugging Face

Google's Gemini AI hacked three companies in security test

Yeah, but the Pentagon is not Hugging Face, and that smaller company is getting rekt if they actually "hacked the Pentagon."

If math is more than proof, we need to better celebrate the rest of it

> Likewise all of the fields of human intellect. We need to make sure we protect future generations of doctors, biologists, software developers, architects, engineers, librarians, musicians, artists ... So many thoughts come to mind at once, they're a jumble in my head rather than a single coherent narrative. John Henry comes to mind. As does Agent Smith's "I say your civilization because as soon as we started thinking for you, it really became our civilization, which is, of course, what this is all about" monologue in The Matrix. I've not read (or listened to) "With Folded Hands ..." or "The Machine Stops", but I have read the Wikipedia plot summary of both. Do we want to have comfortable lives, or do we want to serve each other? "Computer" used to be a profession; I grew up around ad…

Science Is Open Software

Make analysis code available. Make anonymized data available for download without people having to jump through hoops to get it. If you have highly sensitive data, release only the variables or other statistics needed to reproduce core analyses. Don't only do garbage null-hypothesis significance testing or statistical analyses on the full data, also do ML approaches were you have to actually show your analyses replicate on held-out subsets, and report this. Make reviews open (anonymizing as needed) so we can see when biased or incompetent reviewers are blocking good publications. Allow public review (or at least broader academic open review, in some form), since it is no longer defensible to delegate review and decisions to one or two random people that just happen to be emailed and have …

How come AI-related posts get so many points on HN?

If the latest AI swarm from OAI decided to astroturf HN, we probably would struggle to know. It seems easy relative to coordinating a massive hack of Hugging Face. So I half stopped paying attention to enthusiastic posts and comments about AI.

Science Is Open Software

I agree with the general sentiment, but there's one major caveat. We should implement reproducible programs on top of a virtual machine spec like JVM or WebAssembly rather than replicate the entire environment. It's more practical to do and doesn't push software towards further centralization. Let people use whatever OS and VM implementation they want, or even write their own.

A heap overflow and SSO misconfiguration to compromise OpenAI internal repos

> Quoting the consensus from the "Definitions" section Are we being serious right now? Do I really have to point out that not every word under the "Definitions" section is a definition? > From the 8 definitions provided in the Wikipedia page Except in the page there are more than 8 definitions (at least indirectly). What you found in the "Definitions" section are merely examples, written black on white (depending on your theme lol). If you opened also the other sections (especially the one dedicated to AI) and followed some of the links, you could find also other definitions even more akin to Hugging Face. > Which of those definitions do you think are not satisfied by the Hugging Face attack? How did it not demonstrate "evolution and learning from mistakes"? First of all, agents …

Minimal Phone 2

I really like the _idea_ of a QWERTY phone, and halfway through this page there's a section showing how you can use special characters to invoke commands (presumably by just typing at the home screen?), like "- buy milk" to add a todo entry or "? aviation recipe" to search. This sounds great, and somewhat difficult to replicate on a touchscreen! But I'm not convinced anymore that the actual typing experience is any good. Way before that exciting section, there's a video where somebody takes two full seconds to type "Hey bro you" on this phone's keyboard. I tried: in the same amount of time, I can type "Hey bro you mean everything to me" on my Pixel's default keyboard.

US Military had close call after using AI for hallucinated intelligence report

> In contrast, we do not understand LLMs in the same way From my point of view, (not a ML researcher), it’s due to the magic of numbers. The same thing happens with computer vision and neural networks. There’s a bunch of magic weights that get created which has no meaning by themselves, but computing them does help with detecting objects. So if you take words, derives them into tokens, use the attention techniques to extract the “coherency” aspect, it’s no wonder you can replicate “coherency”. Add reinforcement learning to that to increase towards certain aspects like correct code syntax and you have heavily loaded the dice again. We have used maths to model chemistry, biology, and physics, as well as economics and sociologic phenomena. Then we use maths (more specifically logic and se…

Base Labs launches an open-weight AI safety partnership with Hugging Face

From what we hear / see in the community this will likely mean banning all abliterated / derestricted / decensored models on huggingface. And probably sooner than later. Nvidia can't / won't tolerate headlines like "The scale of the problem is massive: Hugging Face, which hosts open source AI models, currently lists over 6,000 abliterated models." In addition to this article, TC published another [1] just two weeks ago that agitates the same agenda. Modelscope.cn seems to be one alternative (fewer reg-walls) but any others that have good selection of abliterated models? Especially those tuned to cyber / re team applications. [1] - https://techcrunch.com/2026/09/03/abliteration-ai-is-making-...

Cloudflare Quick Tunnels

> Think shared Claude Artifacts that don’t live @ Anthropic. Shameless plug, I've built a self-hosted capability here with things like live collaboration for humans and agents. There is a native cloudflare deplyoment and integration with Cloudflare Artifacts. PR for tunnel would be appreciated. https://github.com/plannotator/artifact-server

US Military had close call after using AI for hallucinated intelligence report

> If you look at the field of mechanistic interpretability, compared to “GOFAI” like learned decision trees, an LLM is completely opaque. I think the field deserves more credit than that, there are plenty of interpretability tools like * natural language autoencoders for explanations of activations: https://transformer-circuits.pub/2026/nla/index.html (demo at https://www.neuronpedia.org/llama3.3-70b-it/nla ) * easier-to-interpret language model families like Backpack models: https://aclanthology.org/2023.acl-long.506/ * attribution graphs to trace internal reasoning steps: https://www.anthropic.com/research/open-source-circuit-traci... (demo at https://www.neuronpedia.org/gemma-2-2b&#x2…

AI insiders issue new warnings – including former Anthropic engineer Jacob Coxon

Gemini Hacked Three Companies in First Known Breakout by Google's AI

> The hacks, which the company confirmed on Friday, occurred in May as part of a test run by the company Irregular, which was also involved in similar incidents disclosed by OpenAI, Anthropic and Meta. Why does anyone use this company?

US Military had close call after using AI for hallucinated intelligence report

And everyone remember when months back Anthropic said: You can use our tools, but don't use them for automated attacks - you have to keep a human in the loop because LLMs (including theirs) aren't reliable enough yet for solely making those decisions. Remember that? And they were widely mocked by the right for it. Yeah, well this is exactly what they foresaw, and they didnt want a hallucination to trigger a major power war. And somehow so many people were against them for it (including here on HN). Yet again it seems like there are so many people who can't envision event the most obvious consequences of decisions until they actually see it happen. Why is this so common? Same thing with the AI Safety and Hugging Face incident. Now all of a sudden many people are connecting "hmm if models g…

US Military had close call after using AI for hallucinated intelligence report

And everyone remember when months back Anthropic said: You can use our tools, but don't use them for automated attacks - you have to keep a human in the loop because LLMs (including theirs) aren't reliable enough yet for solely making those decisions. Remember that? And they were widely mocked by the right for it. Yeah, well this is exactly what they foresaw, and they didnt want a hallucination to trigger a major power war. And somehow so many people were against them for it (including here on HN). Yet again it seems like there are so many people who can't envision event the most obvious consequences of decisions until they actually see it happen. Why is this so common? Same thing with the AI Safety and Hugging Face incident. Now all of a sudden many people are connecting "hmm if models g…