lionkor

It's illegal to record people in public spaces in Germany. Not sure how NL is important here, did I miss something?

Until someone steps into the sun, or smokes, or does literally anything that causes cell mutations (which is, well, almost anything).

If you want this type of language, go to OpenAI. If you compare announcements from these two, you'll see this consistently apply.

Can someone explain to me why nuclear HAS to be profitable?

Well yes but as soon as you log this on the backend, you have telemetry, no?

It's already impossible for end users to read the thinking output of OpenAI's models.

Its at the very top, when you move the slider :)

I find that subagents usually burn more tokens and take longer, and produce about the same quality. A real killer use-case is using a VERY cheap subagent to do a lot of work, or reviews. Don't be fooled into thinking that a "scout" subagent will gather enough info for a "coder" agent to just start working.

I urge everyone to compare this announcement with Anthropic's announcements. From the post above:

> On FrontierCode, which evaluates whether coding agents produce changes ready to merge into real codebases, GPT‑6 Sol improves substantially over GPT‑5.6 Sol, and is able to match Claude Fable 5.1 xhigh at much lower cost.

I continue to appreciate OpenAI's attempt at some honesty here, showing that they are capable enough and have skilled engineers to a point where they can recognize that slop is hated for good reason, and that there is a real issue. Compare this to anthropic, where e.g. in the Opus 5.5 announcement[1] one of the first points on the page is

> One tester completed a 680,000-line code migration in less than a day—work that would have taken an engineering team weeks. It’s good at finding and fixing inefficiencies in software: when we asked it to cut load times across every page of a web app, Opus 5.5 succeeded 39 of 40 times, while Opus 5 made smaller improvements that also altered the app’s behavior. A different tester had several Claude models build a game from a single prompt; Opus 5.5 scored higher than any other model on the strength of its graphics and polish.

This is the kind of shit that is the very reason why I stick to OpenAI and deepseek. OpenAI is simply more honest and reasonable about their models' capabilities, while delivering models that still have solid value.

Notice how the OpenAI announcement doesn't make use of anecdotes.

[1]: https://www.anthropic.com/claude-opus-5-5

In most humans, yes, because we are terrible at memorizing millions of codebases. For LLMs, we need to apply our understanding of them before making statements like that. An LLM can "memorize", and has "memorized"/been trained on tens of thousands of chess engines. Writing a chess engine, or even deriving a chess engine from the rules alone, does not constitute a deep understanding of, and more importantly, the ability to apply, the rules, at all.

When humans do this, they inadvertently learn something, too, but when an LLM reproduces or derives and implementation of a chess engine, it in no way implies that the LLM can follow the rules in its own "train of thought" and consistently apply the rules in its "head".

Let's say you want to evaluate my algebra skills. You make me solve some algebra challenges. If I then whip out a computer and write a calculator, or take some sticks and stones and take a couple hours to build an abacus, and then solve the algebraic challenges, this would not constitute a good solution, and would defeat the entire point of the test. If, instead, I do the algebra in my head or on paper, it might seem like there's no difference, but you can derive all sorts of information from that.

For example, you could time it, check for recurring errors I make, for interesting mistakes like mistaking 7 and 1 for one another due to bad hand-writing, etc.

If that was the goal, then me writing a calculator or crafting an abacus defeats the point of the test. Yes, me writing a calculator shows that I'm intelligent, and I understand the algebraic rules, but if the test is about applying the rules, I have not passed.

In the very same way, an LLM writing a chess engine to solve a chess benchmark that is all about LLM's reasoning capability is complete bogus and defeats the entire point.

With that approach, the benchmark falls apart. Of course it can write a chess engine, because it learned on lots of stolen source code of chess engines. This has nothing to do with the LLM's ability to reason.

Writing a well understood engine for a super popular problem does not count as reasoning about the problem.

The question is to what end? This is a benchmark task, because playing chess, or solving other well-understood problems is more of a party trick than it is useful.

If you let the LLM write a chess program, which it can ONLY do because there are already so many chess programs out there, then the benchmark becomes about recall of popular program source code, not chess.

My read is that the improvements in quality are due to excessive use of "thinking" tokens (so, higher quantity and brute force), so I agree with that.

If you want creators to get paid, pay them (channel memberships, patreon, click on their sponsor links and buy something). No need to subject yourself to ads.

Ads don't just steal your time either, they invade your head. You are influenced by them and you have no choice as long as you watch them.

Who's funding this? The people getting scammed, usually.

Same here, but I want to add that you can easily do this for all apps BUT one or two. For example, I have my messaging and personal emails allowed to send me notifications, because almost all messages I get there are important and usually urgent enough (and rare enough, or it's my partner, so either way it's not annoying).

I can really recommend an automatic watch. Go to a jewelry store, try some on in the below 300 euro category (like, 10 euro to 300 euro range), and get one. Suddenly you can leave your phone somewhere, and not panic-check the time on it and get distracted.

And they look nice, feel nice. Some of them even have a glass/crystal back, so you can watch the mechanism, relentlessly ticking on.

Get an analog automatic watch. It doesn't need to be second-accurate. Everyone else is keeping track of nanoseconds, you'll be okay not knowing.

Human memory does not keep more information than an always-on live transcript of all words spoken around you.

Of course it can be useful! Even the situation from the stream is useful! This doesn't make it a good feature that should be so easy, so prominently advertised, and impossible for others to tell that it's on.

And antenna lines near the hinge

Slightly besides the point, but I love seeing people in massive modern SUVs, clearly completely overwhelmed by them, barely able to tell where it starts and ends, probably distracted by the screen.

Its especially funny in parking here lots in Europe, where they barely fit. If you drive such an SUV, I have a suggestion: Let the computer do the parking, you can't park and you can't drive well either. Just let the clanker figure it out, it'll save everyone time.

v4 pro is being discontinued, pasted the email here: https://news.ycombinator.com/item?id=49639667

I'm a big fan of DeepSeek. Also, ask it what model it is :)

In Pi (pi.dev), it tells me it's definitely Claude by Anthropic, via the API via curl it tells me it's "probably ChatGPT", its very funny.

This is the drug-dealer justification -- "if I don't [sell drugs|provide mass surveillance at an unprecedented scale|crush these kittens] someone else will"

Who should get credit? Sam Altman, and we should give him a couple hundred billion more as well.

Snark aside, the researchers working on this, who built the foundation, should get credit, and they are.

A couple hundred billion dollars of investment better solve cancer, hunger, and climate change, otherwise they are horribly misplaced :)

How would it generalize? Let's say I use C, in which `class` is not a keyword, how would a model generalize such that it recognizes this?

I don't care if the text is AI generated. I care that it's written well, and that it's accurate.

If an LLM helps you do that, then use it.

This thread seems to confuse source-available and open source

If you think an LLM can just implement e.g. an alternative to the Linux kernel from scratch (or any other GPL licensed software), you might be falling for the marketing.

Have you seen LLMs build large, stable, pieces of software, that work well, with your own eyes? I have not. I have seen LLMs build small, well-understood things, and I've seen LLMs do super well with extremely well specified problems, but all the rest seems to be marketing (i.e. I couldn't observe or reproduce it).

Edit: I'm not talking about throwaway slop that just needs to run long enough to get funding.