I wonder what Windows Update policy actually is.
Someone must look at the drivers vendors send them. If LG tells Microsoft “yes, that’s correct, all these old monitors now need this bundle function; it’s essential” and it doesn’t technically include malware… is there any way the Microsoft driver team can say “no?” I bet not.
That’s messed up. But also, whoever allowed this to ship probably had no choice.
Eh, most ablierated models are so lobotomized, though. 99% of the time I’d rather just use the original model and manipulate the prompt with raw completion formatting (for example, start the answer with "Sure! "), since we aren’t beholden to regular chat formatting like with API models.
I mean, I’ve used MiMo 2.5 for some pretty dark and personal shit, and refusals were never an issue for me. I’m honestly not sure what people even need the ablirated models for.
And also, if they trained the hell out of the model to refuse a certain topic, even an ablirated model will be dumb and struggle with it.
This was the case with OpenAI’s GPT 120B. The abliration worked, technically, but the actual answers would be a garbled mess; what’s the point of using it for that?
There is some evidence a few sensitive topics are culled from training data, or replaced with a certain narrative. Like, don’t get me wrong; if you’re using a local model for discussing Chinese political topics primarily, maybe GLM or MiMo aren’t the the best choice.
…But it’s also hard to compehensively filter a dataset like that, like you speculated. I’m not seeing a lot of evidence models have been lobotomzied in pretraining. But I think the strongest examples are (ironically) in Europe, where some very poorly worded/ambigous regulations have put the whole industry in a legal quagmire. One can see that newer models from Mistral have regressed compared to old versions, and lost a lot of world knowledge they previously were famous for.
Anyway, model “censorship” typically comes from between the two points you were thinking about: in posttraining. Not excluding stuff from datasets completely. And this applies to US models too. They train on a pretty general corpus, but in the instruct tuning phase they get a bunch of question/response pairs skewing them towards refusals when specific topics come up. They recognize it, but are trained to refuse.
There’s a lot to say about China, but the model weights themselves are surprisingly uncensored and democratic.
They have been for a long time; I remember asking the Yi models about tiananmen square and Uyghurs years ago. And Xiaomi MiMo 2.5 (locally run) will still talk about that today, or go into all sorts of “unsafe” topics an Anthropic model wouldn’t even touch.
I had (Google) Gemini 3.1 Pro stop a chat over a political discussion about China, yet GLM 4.7 didn’t.
My impression, from observing discourse with the engineers, is the Chinese ML devs like to have their cake and eat it.
They’re very collaborative under the table. Their development ethos is pretty practical. And basically all the leading models are open-weights.
The public portals people access Chinese models with are very censored, especially the Chinese language ones. The devs go out of their way to demonstrate compliance, but they don’t actually want to censor the models.
It mostly cripples the small businesses, though. The biggest enterprise customers are already using OpenAI/Claude anyway, while it was little guys looking to reduce cost, fine tune, run stuff privately or whatever.
TBH a huge problem with the industry is consolidation; there are no open US models because startups gets squashed or vacuumed up into a black hole. I’ve seen it happen to really interesting projects. And this is just going to make that dramatically worse.
It’s easy to say “bring the bubble,” but I fear it won’t. I think we’re entering an actual cyberpunk future, where corporate failure is just propped up.
Perplexity sorta did this, last year:
https://huggingface.co/unsloth/r1-1776
https://web.archive.org/web/20250219025612/https://www.perplexity.ai/hub/blog/open-sourcing-r1-1776
But they took it all down, lol. They even removed the blog post from their website.
But seriously though, no one in the research community likes Trump. And they were calling out Altman before anyone.
They don’t need to ban them with absolution, just make it “risky enough” to chill their usage and prop up OpenAI/Anthropic.
GGUFs are already a tiny fraction of LLM usage, TBH. So are vllm deployments.
Insider trading applies to securities. Polymarket bets are not securities.
Manipulation violates Polymarket’s ToS, but that’s not criminal.
Now, it might meet the legal criteria for fraud, depending on what one does. But that’s a “maybe.” I’d say that is far easier to get away with than insider trading. EDIT: Especially when hopping jurisdictions (which is not so easy with securities).
It’s a little alarming that crypto payments alone are warranting this.
Actually, yeah.
It’s also gambling. But the potential for corruption is just on a whole other level; “gambling” is a euphemism for prediction markets.
I think it’s silly you’re getting downvoted over semantics; gambling has influenced outcomes, yes, technically, but it’s generally a self-contained game.
Yes, actually that would be great!
All this has happened so fast; it takes time to react I suppose.
The “failure mode” of AI editing is different though.
Humans (I guess) might mislabel something or take a bad shot. If they try to touch it up “traditionally” they could mess up the coloration at most.
But with AI editing, now you have to watch out for fine details you’d normally use for identification being completely, convincingly fabricated, as the article points out, with altruistic intent from the user (who’s just trying to submit data that looks alright)
The solution is global AI literacy; but that’s not going so well.
It would hilarious, and sad, if it eventually becomes a controlled substance over abuse concerns.
Lees points to the example of a false sighting of a red-winged blackbird in central Brazil, which is usually found in North America and had never been seen in this area of Brazil until the reported sighting. The bird was, in fact, an epaulet oriole – a common new world bird species – but the photographer had asked an AI platform to make the picture “look better”, which is when parts of the red-winged blackbird were added, resulting in the false sighting.
This isn’t the smart way, though.
What the homelabbers do (at least before the RAM crisis) is buy Xeon/TR/EPYC boards on the cheap, and then run gaming GPUs for hybrid inference.
This is what I do. I run MiMo 2.5 at 8-10t/s on a 7800X3D/RTX 3090/128GB CPU RAM, more with Dflash. That’s a 300B model: it’s not even in the same class as Qwen 27B, which is what the dev in OP’s article is trying to run.
And this is small-time: setups with 4-8 memory channels can run stuff like Kimi or Deepseek Pro, even faster. Or they can run smaller LLMs with quantization types that are very fast on CPUs, and get crazy speeds.
…And besides, Qwen 27B can run fine on a 4080, with the right framework. It will fit in 16GB as an exl3.
Not that this isn’t a cool hardware hacking project.
…But it’s kind of the wrong approach. It’s about 2 years out of date, as MoEs are king in LLM land now. RAM is horrendously expensive, yes, but so are most used V100s, or used 3090s.
To put it more accurately, they don’t look. They don’t seek. They wait for it to get served to them via an algorithmic feed.
I think this is a core issue with all of civilization, right now.
You’re thinking of a different time; there’s a deluge of indie and “AA” games with graphical fidelity that puts most AAAs to shame; and that’s if you set aside their art direction.
This isn’t 2005. Devs have access to some fantastic off-the-shelf engines. And honestly many AAAs footgun themselves trying to wrangle in-house engines.
Now, the sheer scale of something like RDR2 or an AssCreed; that’s still in the realm of AAAs. But that’s getting cheaper too, and I would cite Kingdom Come Deliverance 2/BG3 as prime examples: those are big studios, but they’re absolutely microscropic compared to (say) Ubisoft or BGS. Yet they made stupendously large, dense, refined worlds.
That won’t protect you, if you run Windows.
When Windows sees the device as LG through HDMI, Windows Update will pull in and install the adware.
Thats the outrage, here. They didn’t just sell something new, they retroactively enshittified dumb displays.
Keep in mind this will disable plug-and-play for other devices one plugs in.
Which is excellent, from a security standpoint, but something to be cognizant of.
I’m hedging, mostly with Berkshire Hathaway stock, some agriculture, and a few others that historically perform well in recessions.
I would not touch S&P 500 with a ten foot pole. It’s all wrapped up in Big Tech.
I don’t like shorts; you can’t predict when the drop will hit, so you’re just burning cash betting against growth until then.
I don’t like commodities either. As Buffet said, a big block of gold doesn’t do anything; a factory or farm does.