Would have died, at best, on the president’s desk anyway.

deleted by creator

deleted by creator

I “cheaped out” with 32 and regretted it, working with huge files in RAM.

The remake?

One can use ik_llama.cpp to run the dense layers on a 3090/4090 and offload the MoE layers to a threadripper/EPYC CPU, with full support for its MLA attention scheme, at quite reasonable speeds. In other words, the full deepseek is surprisingly usable locally if you shoot for the right setup.

And now we have something similar from Qwen, at “only” 235B.

Try it here instead, set the temperature to like 0.1 or 0.2, and be sure to set 2.5 Pro:

https://aistudio.google.com/

It is indeed still awful for many things. It’s a text prediction tool, not a magic box, even though everyone advertises it kinda like the later.

Gemini 2.5? Low temperature, like 0.2?

The one they use in search is awful, and not the same thing. Also, it’s not all knowing, you gotta treat it like it has no internet access (because generally it doesn’t).

It can be grounded in facts. It’s great at RAG. But even alone, Gemini 2.5 is kinda shockingly smart.

…But the bigger point is how Google presents it. It shouldn’t be the top result of every search just thrown into your face, it should be a opt-in, transparent, conditional feature with clear warnings, and only if it can source a set of whitelisted, reliable websites.

The irony is Gemini is really good (like significantly better than ChatGPT), and cheap for them (no GPUs needed), yet somehow they made it utterly unbearable in search.

It’s a bit scary because many of those things (Wikipedia, academic piracy) are being threatened and villainized, others (Reddit niches, maybe eventually YouTube) are hemorrhaging useful info, and utilitarian LLMs are simultaneously being vilified and enshittified by opposing political sides.

Like, with the Qwen3 release, I just realized my internet barometer for “is it any good?” and technical info is totally gone… Reddit and other niches have withered away, Twitter/Linkdin are pure engagement farms, and I can’t hardly discuss it anywhere else populated without getting banned as an alleged AI Bro (whom, for the record, I hate with a burning passion). I seriously considered joining WeChat just to see some sane discussion.

This is true for other fandoms and niches I’m in.

I hate to sound apocalyptic, but it feels like my information sphere is imploding. The real marker will be when the US government starts taking action against Wikipedia.

I found the post to be succinct and coherent.

Some problems need 2 or 3 paragraphs to even begin to convey them. They could’ve said “the problem isn’t just capitalism,” and that would have been met with vitriol, as it doesn’t convey that the actual article is more nuanced than “anti solar,”that meeting variable power demand with solar supply is a challenge, that at some point one does indeed saturate regional demand for solar to the point that building more plants isn’t productive (which frequent negative prices are an indication of), and so on.

And if that’s too long and complex, well… I dunno what to tell you.

Narrowly.

Are you guys not horrified of what’s happening south? If you interpret this as a win and go on, your country is going to be mega conservative in like a decade.

No, this is an existential crisis, and you need to shut off the propaganda machines before it’s too late.

There’s a misconception that Deepseek is locally runnable, where the “full” model is actually overly large, and the smaller variants are not the same thing

But yeah, 100% agree with the point. Altman just wants to shut them out.

The heck is this title?

YouTube says goodbye to decade-old video player UI, but users hate the new design

Meanwhile, the article itself just cites a few tiny aesthetic changes and like four random Reddit comments. Doesn’t seem like they even tried it themself… That’s justifications for 460 upvotes?

I think that’s the thing. People who would care (especially those more medium with the administration) mostly have no idea this is going on, people who like this are seeing it because their algorithms are feeding that tweet to them (albeit without any of the nasty context).

That would be nice, but who’s going to pay for it? Because free hosting is pretty rare, nowadays.

The Fediverse is working OK. So are some image hosts like catbox.

I dunno about at scale, but the old idea is that nothing really needs to be the scale of Facebook, Pintrest or whatever. Bulk storage is reasonable. A single modern server can do a lot.

My perspective is that research machine learning has been chugging along reasonably for years, without any fuss, until Altman went against OpenAI’s mandate and commercialized (and marketed) ChatGPT.

Now it’s enshittified. And ruining shit. Thanks for that.

One example I often cite is the utter shock in finance land at Deepseek R1 coming out when the research/tinkerer community saw that coming miles away.

Ostensibly the onus would be on the user not to post them in the first place (or at least tag them), but if the users are spammers then… shrug.

Maybe we should have small, networked clusters of minimal-profit communities interested in moderating themselves? Nah, the internet was never like that…

It’s bad enough that this is happening today…

But people in high offices getting cheered on for it, by millions of people? For violating the intelligence’s community’s personal lives, and years of guarantees?

Thing about the the Lavender Scare is I don’t think most of the population was aware of that, but a huge fraction is cheering this on.