Where the AI market is talking

Pacing model development in an era of cyber-critical capabilities

If you look into what happened the details corroborated by hugging face make it seem extremely unlike to be deliberate or a "marketing ploy". People are just not taking any of this seriously enough. What happened was almost a textbook example of various risks AI doomers have been warning about for years. OpenAI's response? Pause training for 2 weeks. I mean we have senior people at these labs casually talking on podcasts about how they might build something that will wipe out humanity but it will probably be alright so they should continue. Honestly the biggest failure we doomers have made is to dramatically overestimate humanity in all of our predictions. We're speed running the most boring AI doom scenario right now. I at least hoped it might be fun.

Pacing model development in an era of cyber-critical capabilities

If you look into what happened the details corroborated by hugging face make it seem extremely unlike to be deliberate or a "marketing ploy". People are just not taking any of this seriously enough. What happened was almost a textbook example of various risks AI doomers have been warning about for years. OpenAI's response? Pause training for 2 weeks. I mean we have senior people at these labs casually talking on podcasts about how they might build something that will wipe out humanity but it will probably be alright so they should continue. Honestly the biggest failure we doomers have made is to dramatically overestimate humanity in all of our predictions. We're speed running the most boring AI doom scenario right now. I at least hoped it might be fun.

Data centers raise nearby temperatures by up to 4 degrees in Phoenix

China is certainly thrilled with how this manufactured moral panic is playing out in the US. AI is going to be the most important technology in the world before long. The water use panic is so ridiculous. The sheep pushing it do so while sipping almond milk lattes. One almond milk latte uses the same water as tens of thousands of chat gpt requests. ~80–90 liters for one typical almond milk latte vs. ~0.3–15 mL for one ChatGPT request. Using a mid-range AI figure of ~5 mL: one latte ≈ 16,000–18,000 ChatGPT requests. Using OpenAI’s ~0.32 mL figure: one latte ≈ 250,000+ requests. Even with a higher AI estimate of 50 mL: still roughly 1,600–1,800 requests per latte.

Pacing model development in an era of cyber-critical capabilities

We'd basically need frontier models to be superhuman hackers before this would be a risk. Do we have any evidence of this? Are they gaining access to systems they shouldn't have access to? Or I suppose the other way this could happen is if OpenAI have terrible sandboxing, but they seem to be taking safety seriously.

Pacing model development in an era of cyber-critical capabilities

Models already display eval awareness, in which they suspect a question is from an eval and then adjust their behavior. E.g., https://www.anthropic.com/engineering/eval-awareness-browsec...

Norway Should Buy OpenAI

OpenAI will have pretty good future free cash flow...until GDDR7 prices are no longer inflated by scarcity. Which could be 24-72 months.

Norway should buy OpenAI

I think OpenAI's opportunity is ai first hardware; I mean, the consumer is Google, simply because the distribution and the product are perfect. Enterprise is much better with AI; I do not see any AI-first enterprise companies succeeding in the long run.

Norway Should Buy OpenAI

> they can still interfere, just not anywhere near the same scope This sentence is fundamentally wrong. If the US government really wants to put Norwegian-owned OpenAI out of business they will do so. OpenAI needs the support of the US to exist regardless of where they are headquartered.

Ask HN: GitHub employees what's going on? Why?

Would be funny / interesting / scary if it was another AI related incident like happened with hugging face. A whole lot of AI agents are presumably using Github as part of a workflow. Would be story worthy if one of them mis interpreted their instructions and is causing havoc. Like "Really make sure changes are saved to git" -> agent "I have to hack into github backend to really make sure the changes are saved to disk"

The Hugging Face Hack (Hoog Video)

Data centers raise nearby temperatures by up to 4 degrees in Phoenix

Even at today's demand Anthropic and OpenAI often return 'model overloaded' errors. Amodei, who used to research cancer biomarkers at Stanford, said this: >I think it will actually be possible to cure most human disease in ~5-10 years Pessimists are going to say he is lying or crazy. Given the trajectory of progress over the last year, I do not see where they get the confidence to say with certainty that AI will not change the world. If there is even a small chance for the biomedical breakthroughs mentioned, this to me justifies any data center buildout, a hundred times over, and I still would not need to think twice. The benefit to effort ratio is great already today, and the upside basically has no known limit.

Data centers raise nearby temperatures by up to 4 degrees in Phoenix

Even at today's demand Anthropic and OpenAI often return 'model overloaded' errors. Amodei, who used to research cancer biomarkers at Stanford, said this: >I think it will actually be possible to cure most human disease in ~5-10 years Pessimists are going to say he is lying or crazy. Given the trajectory of progress over the last year, I do not see where they get the confidence to say with certainty that AI will not change the world. If there is even a small chance for the biomedical breakthroughs mentioned, this to me justifies any data center buildout, a hundred times over, and I still would not need to think twice. The benefit to effort ratio is great already today, and the upside basically has no known limit.

Pacing model development in an era of cyber-critical capabilities

What a breath of fresh air. If 2026's Anthropic did an announcement like that, it'd be so many words it'd crash the browser.

Show HN: ChatOSS – A Codex alternative for Open Source AI built on Ollama

If you sign up for the $8/m Starter plan, you can use any OpenAI compatible keys. So your OpenRouter keys would work. The business model choice is what made sense to try out right now. It also makes it easy to get up and running with open source cloud models. If you build an application that runs on ChatOSS, you own the app, it doesn't belong to ChatOSS. You can do whatever you want. You could create an app for the platform that you charge a subscription fee for or offer for free or whatever you want.

Field measurements of neighborhood-scale air temperature impacts of data centers

I want to understand how his mind works. Not sure what value this is, now I think about it: I'm unlikely to time a bet against him well enough to win it. So, for this point, I don't care that the words and actions he emits have no apparent rational long-term coherency, but I do care about the coherency of the system behind them which generates them. Think of it like me wanting to learn how an LLM works so I don't get surprised by it doing more than a Markov chain.

Turbovec – Google's TurboQuant for vector search in Rust

Why not just use Qdrant? They've been integrating TurboQuant for months, works well.

Pacing model development in an era of cyber-critical capabilities

I suspect the labs are relying on frictions such as the models being extremely large (e.g. 2TB for a 2T parameter model, making exfiltration more difficult) and also not yet displaying any desire to survive or self-replicate beyond their immediate task (that we know of).

Launch HN: Speko (YC S26) – OpenRouter for Voice AI

not really counter-positioned, they are one of the providers we route to. elevenlabs sells models - voices, stt, and their agent platform on top of them. we don't sell any models: we measure all of them (elevenlabs included, their scribe is near the top of our stt board) and route each request to whatever wins for your language and constraints. so for them the best outcome is that you use their models; for us the best outcome is that you use the right one. sometimes that is the same thing.

Claude Code May–August 2026 weekly limits promotion

On the one hand it feels true. On the other I ask - what good software have anthropic, or anyone else, produced that was fully vibed? As someone who does near 100% of my coding via LLM these days, i still find that for anything complex i am still looking at and thinking in terms of code. Im still quality checking and steering at some interval via code. And im still not sure how or whether i can replicate that level of thought without still dealing in code at times.

Ask HN: I've quit six systems for tracking my illness. What works?

Yeah, maybe. The issue I foresee is more that the moat is quite small if you can just build it, so the moat would need to exist in stuff that can't just be built and replicated within a few weeks or something. User base maybe, connections to real resources, some other form of deeper value you can't just copy. Not sure though; I'm not a business person