thorum

I regretfully suspect this actually is the future of UI. You just need smart local models running fast enough to keep the framerate high, but that’s inevitable in the long term, isn’t it?

That’s thinking short term. In the long run there are no economic incentives for humans.

Alternative proposal: Since no one reasonable actually wants a future where humans are irrelevant - including the people in power who generally like being relevant - we could remember that we’re thinking beings with agency and just decide not to use AI to replace ourselves.

We made a new tool. Using the tool for some things makes life better. Using the tool too much would make life worse. So let’s not do that!

Distillation only lets you keep pace with released models. By all indication, OpenAI/Anthropic’s unreleased internal models are so far ahead of Astra/Fable and every other lab that the chance of anyone else catching up is rapidly closing.

It’s not a religious belief. There is a mountain of scientific evidence supporting the idea that ultra processed foods are bad for you, including the study linked in this HN submission, for reasons linked to the biological effect of specific ingredients and processes.

Type any enzyme or emulsifier off an ingredient list into Google and you can easily find a study about its health impact.

That’s true except for this part, which is arguably a bigger deal for the Wikipedia ecosystem:

> Excessive data downloading: Agents we believe to be operated by OpenAI made millions of automated requests to our public APIs to access the knowledge on Wikimedia projects, crawled millions of pages (mainly from our projects Wikidata and Wikimedia Commons), and made hundreds of thousands of data queries to the Wikidata Query Service (WQDS). This traffic may have contributed to a partial outage on WQDS in May.

Even when agents are well-behaved and browsing Wikipedia for ethical reasons, the system wasn’t designed for this kind of load from bots. As OP says, we don’t need to accept this as the new normal.

Not sure what you mean. The top of the linked page is a video showing the product being used. Immediately below are the words “Compose images from scratch” and more examples, followed by “Text-to-image with strong prompt following and a native understanding of composition.” This is all within the first 100 words of content on the page.

In what way?

Sure, but the point is that the labs use more powerful internal models for research work, not public models. Public models tend to lag the internal frontier by a decent margin, and are constrained in other ways by monitoring. It’s just not a useful indicator.

> The researchers asked Anthropic’s Claude Opus 4.8, running on open-source software called OpenClaw

Meanwhile, Navier–Stokes was solved by an internal model significantly more capable than Astra (and therefore more capable than Mythos/Fable).

I’m afraid this sort of experiment is cope. The labs clearly believe RSI is coming soon.

Why does it have to be about that? The author says they agree with the letter in the first paragraph.

FWIW, I flagged and downvoted both of your comments. The article is a very interesting overview of the space and recent developments, and having to scroll past multiple paragraphs about your low attention span and reading difficulties to get to some actual technical discussion was quite frustrating!

It reminds me of the Cognitive Dark Forest hypotheses recently shared here:

> “You are creating your cool streaming platform in your bedroom. Nobody is stopping you, but if you succeed, if you get the signal out, if you are being noticed, the large platform with loads of cash can incorporate your specific innovations simply by throwing compute and capital at the problem. They can generate a variation of your innovation every few days, eventually they will be able to absorb your uniqueness. It’s just cash, and they have more of it than you. So the safest bet again is to stay silent, or at least under the radar. Best bet is to not disrupt - succeed at all … ?”

https://ryelang.org/blog/posts/cognitive-dark-forest/

https://news.ycombinator.com/item?id=47566442

There’s nothing surprising or confusing here.

Outside of tech communities like HN, I see anti-AI rhetoric everywhere. A very large number of humans see AI as a purely evil and parasitic thing, no matter how it’s used. It’s an especially common view among artists and young people.

Whether or not you agree about AI, when you’re communicating to a wider audience, you have to read the room.

Prior discussion from when the service stopped accepting new customers in July: https://news.ycombinator.com/item?id=48803886

When I have some important problem I don’t know how to solve, I of course think about it all the time. Every now and again I’ll think of something that seems extra useful and write it down.

Eventually I’m looking at a whole page of good ideas, and some pattern jumps out at me. I attribute many of my best ideas to this process. Writing lets you work beyond the limits of your own brain, especially the constraints of working memory.

I think that is the primary value in a “second brain” - externalizing your thoughts and reasoning makes them stronger - though the productivity community may use the term to mean some other thing.

Unfortunately once you accept this premise, the next logical conclusion is that the physical world and analog systems are just as vulnerable to exploitation by embodied AI.

Free users have had access to reasoning for a while. o4-mini and the initial GPT-5 launch both included reasoning modes for free users.

They took away the button a few months ago and are now putting it back.

Interesting that all four models converge on such similar designs, for such short prompts.

Interesting read! Creating tests is highlighted as something Claude did well, but it strikes me that all the weaker rejected solutions could have been avoided if it were really good at designig intelligent tests for itself. For example, the first solution “was very specific to the reported bug and wouldn’t have fixed the general case” and the third suggestion “prevented the perfectly valid use of as conversion expressions in go commands as well”. I imagine both of these cases could have been noticed and avoided by the agent if it had planned out adequate tests ahead of time.

The “correct”, elegant way for AI to interact with existing software would take decades and billions of dollars to build. Someone would have to do the hard work of building new APIs, solving decades of accessibility issues, etc.

Or you can show an AI screenshots and ask it where to click.

The team with the most star power and hype tends to attract the best young talent.

If the next big breakthrough in AI comes from Anthropic, good chance it comes from some genius you’ve never heard of who decided to work there because of [famous researcher].

I wish them all the best and hope they succeed, but can’t help but suspect they’ve fallen into deep LLM psychosis. Even if you assume they can build this thing and it works as described and then get past all the regulatory hurdles, the scale of infrastructure they’re talking about is enormous.

I actually think “explore Claude’s understanding of colors” is an interesting concept. A lot of fascinating cultural information gets compressed into LLMs.

Unfortunately for the people mad about this, I predict the only thing they will accomplish by pressuring the rsync maintainers, is to discourage everyone else from responsibly disclosing their use of AI. You’re just going to make people disable Claude attribution on their commits to avoid drama.

Somewhat useless article. To summarize, we have anecdotes suggesting they may work but no one has figured out how to prove or disprove it in a study, and the author has some doubts. Meanwhile supplements can be dangerous if you take too much or have a liver condition, or if you buy them from an unreputable source, as with every other substance on earth. Author confirms it tastes good in milk.

I was really unimpressed by the free Codex (for nodejs/react dev). I think it must be using a less powerful model or they’re limiting it in some other way.

You’re probably right in a literal technical sense, but a very large number of people (maybe most?) would choose “no” if properly informed and asked for consent, and lots of people are morally opposed even in principle to downloading a large AI model onto their computer. I’m not one of them, but they’re out there. So in a cultural sense, it is different.

The models are primitive right now, but we’re clearly heading toward “AI as sound synthesis, human as artist” - much like how producers currently use a DAW to assemble premade loops and sounds from Splice, but with the producer now able to prompt any sound, filter, or effect they can imagine into existence and then rearrange them into a song.

See for example Suno Studio, which is not very good in my opinion, but shows the direction they’re going.