Chu4eeno

No, that's not how anything works.

You can finetune an LLM to a new tokenizer by nudging just a few layers (even wildly different kinds of tokens), and there's nothing stopping a lab from using someone else's tokenizer.

Something seems to 404:

    GET https://rorz.io/_next/static/chunks/link-CKkD60iF.js 404
    Loading module from “https://rorz.io/_next/static/chunks/link-CKkD60iF.js” was blocked because of a disallowed MIME type (“text/html”).
    TypeError: error loading dynamically imported module: https://rorz.io/_next/static/chunks/link-CKkD60iF.js index-ConxhaY_.js:2:49967
    The above error occurred in a React component:
    Lazy@unknown:0:0
    div@unknown:0:0
    div@unknown:0:0
    div@unknown:0:0
    body@unknown:0:0
    html@unknown:0:0
    Eo@https://rorz.io/_next/static/chunks/vinext-B9_QPQZC.js:32:16818
    [...]

Don't mind the web designers calling themselves engineers.

There's a lot of annoying issues with Python, but compared to the billions of dollars and thousands of man hours that has been spent trying to fix Javascript and how horrible it still is, it's a perfectly cromulent language.

Probably that british professor who killed himself after getting exposed for fraud.

Yes, it was invented by an anonymous user on 4chan in the early 2020s, then sanitized by a Google paper some years later as CoT (chain of thought) or thinking tags, where the model iterates on its own before writing a final response.

All models do it now.

No, they are correct, residential proxies behind a big ISP's CGNAT is much more valuable for miscreants because you can't block the individual who installed a free android flashlight with a residential proxy SDK embedded.

But the overall latency will be longer than expected because you add the latency from the proxy connection as well. If you're able to ping back to the connecting IP in combination with some clientside js to help measure total latency I think you should be able to reliably detect proxy users in general?

> NixOS has a longstanding issue where library loading (and therefore program start, particularly GUI programs that load a lot) is n^2, but programs still all start in ~1 second or less on my n100 minipc. I still found that to be annoying enough to look into why it seemed so slow (and learned about the loading issue). Can't imagine things taking so long that you actually switch to another task. The same minipc can easily handle my 4k 240hz HDR monitor as well, and the whole experience is completely smooth.

KDE actually solved that long time ago with kdeinit (https://web.archive.org/web/20010517071320/www.suse.de/~bast... and follow-up work by Lubos, see e. g. https://conference2005.kde.org/slides/kde-performance/index....), but the new crop of developers didn't notice much difference on their high-end systems with SSDs, so they ripped it out.

It sounds like Wayland.

When you decide to suddenly stuff some of the most complex pieces of software together (the X server and the compositor which sometimes also double as a desktop shell) into giant codebase/process with seemingly no proper plan or roadmap and make up the specification along the way (or just don't, wayland still has no proper XACE replacement, so kiosk tooling etc. is ad-hoc), you get Wayland in its current form.

Rant over; if you want to actually find out what is making stuff stall, maybe try bpftrace, DTrace or SystemTap (depending on what your distro ships and what tooling is more familiar/suited to you)? Or just run `perf top` and see if anything stands out.

For the boot run `sudo systemd-analyze critical-chain` and see what it is waiting on. And be ready for the nightmare that is the systemd "design": https://blog.darknedgy.net/technology/2020/05/02/0/

Arch actually has testing repos (and has had for as long as I can remember), plus other project-specific testing repos like kde-unstable when upstream is taking a very long time to push out a new actually stable release. It also has lts packages for projects that offer those (like the kernel), and sometimes stays on old versions and backports fixes if the upstream definition of stable is too out of whack.

People seem to believe arch is much more yolo than it actually is (though forks of arch have tended to be pretty yolo, and I assume cachyos is a bit yolo as well based on the issues people have).

It has to be a pain, everything depends on them (or rather, should, you still have holdouts like go fucking up things like NSS, holding back posix_spawn etc.).

In their defense the people working on glibc and gcc are pretty intelligent and spend a lot of time thinking about how everything they do affects their users, though they make mistakes like all humans (and LLMs I should say by now).

Last I tried that (because everyone claimed it was magically better at everything) sound just randomly broke (I had no idea why, debugging pipewire is a nightmare), and since I switched back to pulseaudio everything just works again (and seems to use less CPU, but that could be selective memory).

Maybe I'm just old and grumpy, but we spent decades hammering out all the incorrect (for pulseaudio) assumptions in every layer of the stack and papered over impedance mismatches, and then people decide to rewrite from scratch again (though I'll admit I was a vocal pro-dmix anti-pulseaudio hater for a long time).

Norway seems like an obvious counter example? If the unions and state grow enough power before resources start to get extracted it seems like they can keep the curse in check.

It has also been protecting politics from capital in various ways; public funding of political parties, strict bans on various forms of political adverts, forced disclosure laws, essentially forcing capital into a single visible chair (NHO), etc.

OWS showed that you don't even need drones and robots.

Try looking at the SVG code for one of those converted from bitmaps, and I think you'll understand why even a massive LLM has trouble generating it on the fly (the pelican SVGs are always one-shot single API calls, so no iterations and verifications).

I've written more than a couple of "slide rendering engines" (though usually with an actual renderer, not a website templating thing like reveal.js).

> Second, I want to emphasize and unequivocally clarify that every single winning submission went through at least 2 human judges, and in some cases, up to 3-4 human judges. These judges reviewed and scored the submissions independently based on the rubric we highlighted on the hackathon page.

How did you verify this? The results seem to indicate otherwise.

Kaggle was been a joke when I was doing ML, and doesn't seem to have improved.

It was a neat place to host/download some big datasets before huggingface.

> A good test is to try to port significant from c++ to plain and simple C using "AI". Maybe it can retain some of the semantics for the original code.

So updating cfront for modern C++.

Anathem was so chock full of interesting ideas, too bad the last parts of it were so rushed (or at least felt that way to me).

It smells like Gemini to high heavens, how familiar are you with its writing? If nothing else because of the complete lack of relevant links.

One of the issues with Wayland is that it forces every compositor to reimplement the X server, badly.

So you can't really "fix it", short of patching e. g. Gnome's compositor, with patches that will never be accepted upstream.

It's quite easy to get it to do syscalls directly, I even got Fable to do a simple standalone TLS implementation (in C). Not really useful since it hardcoded a set of supported algorithms etc., but it's fun to see how "simple" you can force it to go.

Evidence in the other direction (that they're able to generalize) is that I can't think of any LLM currently that can't create usable (placeholder) SVG icons, I tried a bit before the pelican became popular and it was abysmal.

SVG output is useful, though. I often ask whatever LLM I have open to generate placeholder icons whenever I need them.

People used to anthropomorphize pet rocks, so people either refusing to be nice to LLMs or getting upset when people anthropomorphize them says more about those critics than anything else IMHO.

Humans are social and the ELIZA effect is perfectly normal and expected imho.

> The EU was good for farmers, not so good for fisheries.

It was good for industrial farming, not the farmers themselves, one of the big reasons for Norway refusing to join was the impact it would have on farmers.

You are very naive about the neutrality of Wikipedia.

Maintaining the Wikipedia page is part of what any PR/reputation management agency does.

I think you have rosy eyed glasses (or never inspected the output too closely), Gemini 3.1 Pro was very bad at hallucinating.

wikipedia happily censors domains, it's not a free speech community.