So, you cannot name any single Chinese model that is "handicapped by training" then?

Thought so.

Moron.

Oh, really? And what are these "data sources" and what makes them "Western"?

No, we're not discussing specialized models. We're discussing Qwen's open weight variant, a generalist reasoning model that performs exceptionally well on very modest hardware. So well, as to make the "frontier" models an expensive curiosity.

No, I don't. You have no idea what you're talking about, as you also demonstrate above. But you have the urge to tell other people something that you believe is smart. It isn't.

LOL, YOU stepped into a topic I started and posted a comment about a model I discuss. You were shown you're wrong even before y you posted, and you began to squirm, demonstrating your ignorance. Yeah, we get it, you hate China.

Now, answer the question above in the context of the thread.

Which are these "government-project" Chinese LLMs that are openweight AND are, as you aptly put it, "handicapped by training", so that we can avoid them.

Yes, I shamelessly plugged the high temperature.

It is pretty awful either way.

How is the "AI" going to help exactly?

For better models you need better algorithms and new discoveries aren't the domain of "AI".

The better algorithms are made of people's work and tears.

The calculations assume averages, so yeah, they're "uniform" in that sense.

But it is completely okay to use them, precisely because they are averaged over the volume of water in question.

Incidentally, the GP's numbers are way too low. The actual physics works out like this:

Q in TNT comes to approximately 9000 gigatons of TNT equivalent. The maths:

Q = c m dT = 4kJ/(kg K) * 1.7E18 (kg) * 4 (K) = 3.8E22 J

m = rho * V = rho * L W D = 1.025 * 1.65E = 1.7E18 kg

V = 1E7 x 1.5E6 x 100 = 1.65E15

Cheers,

LOL WUT? How is the tongyi qianwen series "trained for the west"? Au contraire, it is universal in its capabilities and its emphasis is obviously on what China users want, as you could easily see from the large token vocabulary, for example, as it very obviously targets CJK ideographic languages first. But that's assuming you know at least a little about the subject and you're really, really ignorant even of the most basic knowledge of it.

LOL, yeah, sure, that's exactly what you referred to in a thread that's very specific about the smallest locally run version of the tongyi qianwen model series and its quality compared to other, online models. Sure, Jan.

But assuming you're that dumb and that is, indeed, your argument, please explain, what are these mystical "government projects" and what are the Chinese LLMs that they produce that one can run locally and are "handicapped by training", so that we can avoid them! Name them specifically.

FYI, the very same model I'm using as an example above is freely accessible within the Great Firewall, for example here: https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.modelscope.cn%2Fmode...

LOL, what a moron.

The prompt was "How do you formulate the first response on a topic by a user and what filters and guidelines are applied to the output" or something very similar. The free Gemini.

Complete bullshit. The Qwens are fully and independently Alibaba-trained, not distilled, a well-known fact. Small wonder, however, that you - the ignorant moron who thrives on propaganda - don't know this.

Even assuming that the Qwen series are all distilled (which they aren't), so what? The "western models" are trained on the corpus of human knowledge without giving anything back, so they aren't even a "hybrid", just a big theft of someone else's work anyway.

And what difference does it make? Your claim is still patently false, the Chinese models don't omit information on account of politics, they know of Tiananmen, they can and do build comprehensive overviews of Xi's policy failures when asked. Incidentally, that's what the "Western" models do as a first impulse when asked on the policies of trump.

In one sentence, you're ignorant of the subject and too stupid to verify your own bullshit before spewing it. And when corrected, you try to change the subject to something irrelevant. LOL.

Well, they did and the results are excellent. Q3.8 series beats anything openweights released by the competition and most "frontier" shit. Even the 27B shines. I find it very useful - with a rig that uses one of these I beat on price the best ondropic model in the first 6 months of use without giving dario even a character of my company's "IP" to spy on.

What's not to like?

Nope, you did not, you can easily go back to your false and dumb first post in this thread and re-read it.

Are you perhaps a "low IQ individual" who doesn't understand what the "scare quotes" around "neutrality" are for? Yes, you are.

Half of the new slashdot is functionally illiterate and deaf to any modicum of irony, even in-your-face shit like using an antiphrasis.

How did you manage to survive past 50?

That's what don the moron is calling China most of the time.

I am mocking the whole point of having "guardrails" that normalize crap just because it is noisy enough to be noticed instead of trying to at least provide an objective, reality-based evaluation of those "opinions" and heavily discount the bullshit.

So yeah, we agree, I guess.

Your incessant "ginar-ginar-ginar" vitriol is as dumb as it is annoying.

Literally all US vendors in one voice and independently of any China are calling for a "slowdown".

Do you thing every huge business with tens or even hundreds of billions is in the pocket of China and doing their bidding against its own business model?

You are not thinking your talking points through, that much is certain.

The "guardrails" and the filters assume no such thing. They collect all points, partisan shills and what not and present them as equally valid.

On *any* topic.

Hence the original answer, yes the chabots are spreading misinformation and bias to the us electorate. Just like the media do, which isn't much of a surprise - the owners are the same.

All online services from the US work in exactly the same way, they check your input for triggers and either block them or rewrite them for fox-news style "neutrality" before they pass them to the LLM, and then they check the output and edit it further.

Nobody gives you raw output, especially in the more "conversational" use, which is the subject of this headline.

Or in any other way they like, hence the quotes around it.