c7b

Frankly, that's the one part I would have done differently. Mixing font sizes rarely looks good, but that's just my personal opinion. If you want to visually indicate intensity, I'd have used color, or rather saturation. Not the text itself, maybe a small pill next to it. But I don't think it's necessary, the order already conveys some sort of ranking, I don't think cardinal information adds much here.

Congratulations! I was going to comment on the scroll field in particular when I saw this. I didn't even realize you had to hand-craft the component, but it's such a nice UI idea in general, the way the scrolling works and how the content above changes.

You must be referring to SEPA Direct Debit, which would be very risky to use to siphon funds. There's an 8 week no-questions-asked refund policy, 13 months for unauthorized transactions, if the payment fails (eg due to insufficient balance), the payee is charged a non-refundable cancelation fee. [0] And that's all separate from any legal troubles for fraudulent charges. I don't know why Germans use PayPal over SEPA, but I'd be surprised if SEPA Direct Debit was the reason.

[0] https://www.europeanpaymentscouncil.eu/what-we-do/sepa-direc...

If you're mad at Bluesky for doing this, I think you're mad at the wrong party. Modifying intent is problematic and can be abused (although I think there's nothing particularly problematic about this particular instance). The real problem is that you have next to no possibility of modifying the behavior. On an OS that respects user freedom, there would by a myriad ways of intercepting what the software is doing to get your desired result. On iOS, you're at the mercy of Apple. I think we should be mad at Apple for normalizing a computing culture that views user freedom as a security risk, potentially something that should be outlawed (thinking of age verification).

It's a noble cause, but there are probably bigger levers to pull than the model size if you care about environmental impact.

If you're running Qwen3.8-27B on energy-efficient hardware like a Mac or a DGX Spark instead of an API (likely running on H100s), I'm sure you're having much more of an impact than you would by switching to, say, a 9B coding-only model on the same hardware. The thing is, I think you won't be able to go orders of magnitude smaller, because a lot of the usefulness of LLMs comes from emergent smartness, and you typically need a minimum amount of complexity to see such emergent phenomena (and I think we're pretty far from understanding this kind of emergence, much further than from the next model generation that annihilates the current one on benchmarks yet again).

> I understand that to mean that they are the same architecture, just one version has a massive training set and the other has a very small subset.

No. It means that the one model has 2.4 trillion parameters while the other has only 27 billion. I don't know the details about their architecture or training, but presumably they used the same or similar training sets for both and a conceptually similar architecture, scaled down. I'd guess they also have some techniques to re-use some of the work done for the big model for the smaller versions (if anyone knows more about this I'd be interested). The architectures cannot be identical by definition because then the parameter count would be the same. Subsetting the data to such narrow fields as you describe could risk losing some edge, there are a lot of emergent capabilities in those models and I don't think that emergence is fully understood yet. There are subject-specific models, but for far broader subject areas than you suggested, like coding or math or prose.

I'm sure composability is possible in principle, I'm just sceptical that it'll be a good long-term solution, for my originally stated reason. It's basically just The Bitter Lesson again, we may gain some short-lived edge by putting more domain knowledge into the algorithm, but ultimately (these days often: surprisingly quickly) it'll be outgunned by something that just leverages raw computation better.

At a more technical level, what do you suggest? Training a small LLM on Python code exclusively? And then one on general CS/algorithms, which you'll also need? I don't think the current transformer architectures would compose as you suggest.

Sounds a bit like 'I want to make horses faster, surely I won't need mechanical engineering knowledge'. We don't know everything that we don't know, so it's hard to say what we don't need to know.

I've been wondering whether that is a feature of the foundation model or whatever finetuning they do on top. I remember this from the earliest versions of (pre Chat-) GPT I've been using, which would suggest it's a feature of the foundation model. But I don't really understand why. Something that's been trained on StackOverflow and BB forums, among other things, should have seen a ton of examples of answer refusals.

For those commenting on the long reasoning, it may be interesting to know that the reasoning effort is set to xhigh by default [0]. Other possible values are medium, low and none. Flag for changing it in llama.cpp below, but note that the long reasoning seems to contribute a great deal to the quality.

  --chat-template-kwargs '{"preserve_thinking":true,"reasoning_effort":"medium"}'

[0] https://unsloth.ai/docs/models/qwen3.8#thinking--preserve-th...

The way push notifications work today is a result of said optimization. Smartphones didn't start out with constant notifications, nor did any major desktop OS have them built in the way they do today.

Not sure if that makes it better or worse.

There is some utility to ads. Knowing that my favorite shop has a promotion, that a movie, game or book I've been waiting for is out, or something other that might interest me has some usefulness. The problem isn't the ad per se, it's optimization through A/B-tests and the like, mass psychological experiments with all the rigor of scientific studies and none of the academic ethical safeguards for human subject experimentation.

Fwiw, afaik in most Oxford and Cambridge degrees the only exams that count towards your grade are in the final year or the final two years. Everything before that is basically preparation, and the colleges doing weekly supervision and regular mock exams to keep tabs who is on track to pass and who needs extra attention. I've never heard there was a problem with a lack of rigor or recognition for the degrees awarded there. So U of Michigan taking a small step in that direction shouldn't be the end of the world either.

My gut feeling would be that most of not all of the points apply to Rust as well, and even more so. Rust's compiler is famously strict, the whole language and tooling is designed to catch footguns early, and it's less verbose than Go, meaning you can fit larger codebases into a given context window.

I think the educational policies are most interesting to compare. There was systematic attrition of academic talent driven by xenophobic ideologies. It started slowly and well before the 1930's, before it turned into an all-out purge. There was also ideological alignment of academia in multiple ways. The timing might not perfectly line up with what we're discussing now, but it definitely qualifies as within-lifetime, and I think the overall parallels are interesting.

I would agree with that stance and that classical antiquity ended then. But you still have to acknowledge that the Roman state, its culture and institutions not only survived but arguably came out even more dominant (with the exception of pagan religion), e.g. with a second political stronghold in the East.

There are many episodes in Roman history that people would have experienced as decline or even collapse within their lifetimes, including multiple sackings of the city of Rome itself. But in most of those cases things actually did bounce back and the Empire continued to dominate. That's kind of the point I was making. Empires would seem to be far more resilient than contemporary observers would think if one only looked at Roman history. But the problem is that the Roman Empire is a bit of an outlier in that regard. Ancient cultures of Central America would be interesting to study there (there were far more than just the Mayas, Inkas and Aztecs, most of them relatively short-lived, hence good case studies for the question of collapse). Unfortunately much less is known about them than about Romans.

The decline of the Roman Empire, depending on how you define it, played out over centuries, if not a millennium. It is such a bad analogy for the kind of within-our-lifetimes collapse that is being speculated about here that it undermines the original concern. The decline of Germany as the leading scientific powerhouse in the first half if the 20th century would be a better case study imho.

Seems that this is a patent application from March, so a challenge should still be possible. But it would have to come from a named entity afaik (not a lawyer).

But then how would you get into the news for how dangerously good your models are? And have something to warn about how dangerous open weights models could be?

> opting for authoritarianism before eventually the center can't hold any more and it all collapses in on itself

It all depends on what you define as the collapse of the Roman Empire, but the earliest sensible candidate date would be about four and a half centuries after the population 'opted' for authoritarianism (which wasn't much of a choice for anyone btw).

People need to stop making senseless historical comparisons.

Hiring people isn't retirement, if anything it means growing your business. You're outsourcing some of the menial tasks and inherit managerial tasks instead. The show and the ROI on all those expenses still ultimately depend on him.

That's a completely normal evolution for any succesful influencer. In what world is that retirement? The only thing I can see being retired is his old online persona, the new persona seems more mature, likely tailored to please his fanbase who presumably have grown out of their teenage instincts (and have more money to spend now).

And what does taking it seriously entail?

I keep hearing he's retired, yet his channel looks just like any channel by a full-time influencer. The few videos of him I've watched (all quite recent) must have required substantial effort. Like, he's showing off extensive Linux ricing, complex UI/orchestration tools for LLMs,... on top of refined video editing. Whoever actually made all of that, that must have required non-negligible resources. Probably a lot more than videos in his older style, which were more like stream recordings. The cadence of his videos also seemed comparable to other channels I follow. And his videos were all monetized hard, through YouTube, in-video sponsorships, and even in the video description he was selling his subscription newsletter. To me this seems more like he's still very much in the business of selling online content and just evolved his style.

tl;dr: I don't think you and I have the same meaning associated with the word 'retired'.

I think those concerned about ensuring a place for human mathematicians usually go in different directions than my suggestion, at least those I've seen so far. Like this post that was recently featured on HN: https://kirwinhampshire.substack.com/p/the-dark-night-of-mat...

My perspective is more like a FOSS philosophy for math. Even if a closed version has the same immediate effect, it's just better for everyone if everyone can look under the hood and tinker with it.

I agree with your reading of the presentation and I mostly agree with the presentation - but I believe the recommendations should go a bit further than they do there.

I know it sounds unrealistic and not aligned with academic incentive structures. But those are the exact structures that gave us a lot of headaches in the experimental sciences. I think it would be a good north star to aim for something that resembles how those are trying to address the reproducibility crisis. Better than to embrace the most black-box version of math that AI systems can produce (million-line proofs without context). Even if a reproducibility crisis is seemingly impossible (although agents so far have also been pretty good at finding compiler bugs).

Because the math isn't solely about the proof being correct. You don't need to take my word for it, here's one of the most famous living mathematicians' take on it: https://teorth.github.io/tao-web/slides/age-of-ai-icm-2026.p...

I believe we're seeing a new kind of mathematics that will require completely new formats for publication, a bit similar to those used in experimental sciences. AI-powered mathematics should be fully reproducible, so it's the authors' responsibility to disclose the exact model type, inference settings/seeds and the full prompt history leading to the result. Of course that would ideally require open weights models.

It's not just about requiring to disclose AI use. AI-powered mathematics is a completely valid discipline that doesn't need to be shy, but it should develop its own publication culture.

I haven't driven a recent model but afaik the answer for most is Yes, or at best you can disable them for the current ride and they turn on again automatically on every start.