HNHacker News
TopNewBestAskShowJobs

jstummbillig

6,608 karma · joined February 14, 2015

submissionscomments
jstummbillig··on Creatine raise brain energy levels and slow Alzheimer's cognitive decline by 30%
Where does the 240 number come from? Can anyone tell?
jstummbillig··on Creatine raises brain energy levels and slows cognitive decline: study
Sounds like something that should fly under a different headline until then.
jstummbillig··on Danish pension fund excludes SpaceX citing governance and valuation
> I really don't want my investments to be in SpaceX, OpenAI or Anthropic.

This just being an incomplete list or is there another reason you name the last two but not Google?

jstummbillig··on Claude Opus 4.8
Including for DeepSeek you mean.
jstummbillig··on Five frontier LLMs disagree on 67% of 1k real-world fact-check claims
I don't see how something being lazy for a decade makes it any less lazy. And lazy still seems right to me: They make a misleading point by omitting to collect and present important data. If the headline read "LLMs disagree on 67%, humans disagree on 75%" it would clearly project something very different.

Granted, there certainly are other unflattering adjectives one could have chosen to describe this instead.

jstummbillig··on Five frontier LLMs disagree on 67% of 1k real-world fact-check claims
It's all fairly lazy to a degree that is mildly confusing. I also feel this among other issues would have become obvious if they had bothered to include a human fact checker baseline (i.e. asked multiple human fact checkers the same questions).
jstummbillig··on YouTube to automatically label AI-generated videos
> Something went wrong with music and culture in recent times. Participation became consumption.

I think closer to truth is: Participation became production.

More people are doing more things (including with instruments) but often times in a digital setting, sometimes more isolated and sometimes much more public (think: Twitch streams where chat is part of the whole social experience in a way that was never true for TV or other live events of that scale). More participatory online and more individualized as consumption, while some older forms of face-to-face amateur participation have become less socially normal or less visible.

This says not so much about music or culture really; it seems fairly aligned with where our lives and how we connect have moved more broadly.

jstummbillig··on Where does next-token prediction leave us?
People also don't make data centers, GPU, or do AI research collectively. The part that is made, collectively, are the layers of knowledge required to provide either.

> Electricity has almost always been a public service.

I have no clue where that would be true, if anywhere. Not where I live. I am paying a private company, and the electricity that they are selling me is generated by other private companies.

jstummbillig··on DuckDuckGo search saw 28% more visits after Google said people love AI mode
Why would that be problematic? People are AI outraged. Some of them will move. Those who stay like it better.
jstummbillig··on I think Anthropic and OpenAI have found product-market fit
> 200m knowledge workers in the world, 30m developers. We're talking about a world where you need 5% of every knowledge workers salary to go into tokens. 20% if you're a developer.

This is where the napkin math is breaking down in a big way. There is absolutely no reason to assume this will only impact "knowledge workers". Farmers use computers. Farmers will use AI.

jstummbillig··on Where does next-token prediction leave us?
So, where does electricity leave us? In a perpetual loop of rent-seeking for something made with humanity’s collective output of centuries. It is not a good place for an individual to be in, regardless of class.
jstummbillig··on Greg Brockman interview [video]
I mean this in the best sense possible: Ilya is obviously a very strange human being. People seem to always entirely ignore this when discussing what happened at OpenAI and I feel including it serves well when trying to explain much of what happened.
jstummbillig··on Where does next-token prediction leave us?
> So, where does next token prediction leave us? In a perpetual loop of rent-seeking for something made with humanity’s collective output of centuries. It is not a good place for an individual to be in, regardless of class.

In what way is this different to electricity?

jstummbillig··on Claude is not your architect. Stop letting it pretend
> The flip side of this problem is that it is also easy to phrase prompt in a way that invites _too much_ criticism, so you wind up sycophantic in the other direction where the completion rejects a perfectly good idea because the prompt leads a little bit in that direction.

I don't think that is the flip side. That's just obviously bad. Everything that is obviously bad, the model makers will also ~notice and work to make better. They seem to be a competent and attentive bunch, on the whole.

jstummbillig··on Greg Brockman interview [video]
Granola notes are a 1 Minute read: https://notes.granola.ai/d/2c35c84f-6eb4-497a-8419-294d92141...
jstummbillig··on I spent 50 hours drawing a line graph
> A professional draftsman of the 1920's may cringe at the imperfections in my line graph above. They can suck it.

I am willing to suck it but the kerning is still killing me. (I love everything about this btw)

jstummbillig··on Microsoft starts canceling Claude Code licenses
I think it's great. People at a broad scale are getting first hand experience with resource management. It's a fairly cheap way of doing it too (in contrast to: learning this by managing humans) and we can all benefit from the skill transfer.
jstummbillig··on Project Glasswing: An Initial Update
Ngl, watching folks getting irritated about normal employer-employee absurdities from the employer perspective through usage of agents and having to pay for tokens has been a little therapeutic for me.
jstummbillig··on Lost Images from the 1945 Trinity Nuclear Test Restored
Increasingly with more powerful and precise technology the relative danger of the bomb will decrease, to the point where the bombs will become meaningless. The stigma is too big, the blowback is too severe and we have countless other ways to pummel each other into oblivion with more finesse.
jstummbillig··on OpenAI to confidentially file for IPO as soon as Friday
Given the nature of private conversations, I suppose there is no source to this claim?
jstummbillig··on Anthropic is expanding to Colossus2. Will use GB200
Why are they selling compute instead of using it to build that SOTA model?
jstummbillig··on Gemini 3.5 Flash
> Interesting pricing direction.

Is it? More capability, more demand, higher price. Seems relatively uninteresting. The naming structure complicates it: 3.5 Flash is less comparable to 3.0 Flash than it is to 3.0 Pro.

More generally, $/token + naming scheme comparisons are just confusing: I am not looking for a wordy idiot and I doubt most people are (at least not with what I would consider worthwhile business ambitions). In fact wordy idiots are fairly costly, because we have to consider the large amounts of cheap garbage that they are producing, and if you price your own time somewhat competitively then fairly quickly that's the bigger lever.

Even if we don't consider the last part: How do we price the better model, that can one shot a task without having to go back and forth and spending more tokens or having to fix more bugs later? It is definitely worth something and I think it's quite undervalued right now. What seems to be missing is a better measurement of capability per token. I don't know how that could look like. Maybe something like how we try and measure inflation, some basket of tasks (which then ends up being part of the training data so idk).

jstummbillig··on There's no earthly way of knowing which direction we are going
Contrary to the chorus, I find this mad episode in our history has been going surprisingly well. I mean, it's a lot to process and the speed is ludicrous. What did we expect? What is the baseline? 80% approval in face of unprecedented uncertainty?

We are adjusting, we talk about things. I hope we will keep doing that.

jstummbillig··on Gemini CLI will stop working from June 18, 2026
I would wager your feelings conflict with their analysis (which they are probably quite good at, with all that practice)
jstummbillig··on Google changes its search box
> when instead it's just aggregating almost random stuff

How do you know that?

Scraping websites is literally what Google does best, stringing together information in the pattern of "some people x, other people y" requires 0 AI and could have been done since forever. I find it implausible that otherwise obviously capable models would be reduced to do something akin to just that.

jstummbillig··on I’ve joined Anthropic
2-3 years is pretty average tenure inside the EU tech sector for the past few years [1], but regardless I don't know what that tells us, given that nothing else about this is average. The sample size of Andrej-sized talent in an ongoing tech revolution of epic proportions is just very small.

[1] https://ravio.com/blog/employee-tenure-trends

jstummbillig··on Cursor Introduces Composer 2.5
1. GPT 4 has gotten 6x cheaper over it's evolution (from initial release to Turbo to 4o). Maybe you meant "Only since 4o and only since its final release". Alas.

2. We are not interested in how different model naming schemes relate to prices, we are interested in the capabilities. So if you want to learn something about price development you need comparative levels of capabilities, and then look at the prices. 4o is not comparable to 5.5 in the first regard. It is (according to the benchmarks) maybe more comparable to current 5 nano - which is 98% cheaper.

jstummbillig··on Cursor Introduces Composer 2.5
Isn't there a cli version of cursor by now?
jstummbillig··on Mistral's CEO: Europe has 2 years to stop becoming America's AI 'vassal state'
No
jstummbillig··on Mistral's CEO: Europe has 2 years to stop becoming America's AI 'vassal state'
The trajectory all models are taking, while software development requirements continue to stay fairly level – at least I don't see why they wouldn't. We will still mostly need reliable, boring software.

I don't see why Mistral would stop getting better. It will, just more slowly. Eventually it will be good enough, and continue to be good enough.

← PreviousPage 12 of 34Next →