China Is Shipping More Open AI Models Than US Rivals as Tech Competition Shifts (msn.com)
- Reference: 0179771642
- News link: https://slashdot.org/story/25/10/13/1719227/china-is-shipping-more-open-ai-models-than-us-rivals-as-tech-competition-shifts
- Source link: https://www.msn.com/en-us/news/technology/china-leads-the-us-on-this-measure-of-technological-influence/ar-AA1OmAwI
Irene Solaiman is chief policy officer at Hugging Face. She said Chinese companies build their user base by shipping frequently and quickly. American companies like OpenAI and Google keep their best models proprietary. Meta once led in open AI models. Mark Zuckerberg argued last year that the world would benefit if AI companies shared their technology freely. He pledged Meta would release its AI openly. The company has [2]since become more cautious . Zuckerberg wrote in a new essay that Meta might need to keep the best models for itself.
[1] https://www.msn.com/en-us/news/technology/china-leads-the-us-on-this-measure-of-technological-influence/ar-AA1OmAwI
[2] https://meta.slashdot.org/story/25/07/14/2048202/metas-superintelligence-lab-considers-shift-to-closed-ai-model
How is this even "tech" anymore? (Score:1)
AI solves no real problem, no one asked for this, it gobbles endless resources, produces slop for the feeble-minded.
Re: (Score:3)
Wrong
While it's true that pop culture AI produces "slop for the feeble-minded", serious scientists and engineers use the tools to produce real, useful results
Re: How is this even "tech" anymore? (Score:2)
Show me.
Re: (Score:2)
One example is [1]AlphaFold [wikipedia.org] an AI program which predicts folded protein structures "with near experimental accuracy" from amino acid base sequences. This ability is going to have a huge impact on many practical problems like pharmaceutical development, agricultural science, and engineering custom proteins. For example, since the human genome has been long since sequenced, the program means we now, with a fairly high degree of certainty, know what all the protein coding sequences make.
I'd say that's a pre
[1] https://en.wikipedia.org/wiki/AlphaFold
Re: (Score:2)
> serious scientists and engineers use the tools to produce real, useful results
They do. For example, that "AI" adds tons of security bugs and makes coders less productive. Nice engineering research.
Oh, you meant directly for engineering? Not so much. That is all smoke and mirrors.
Re: (Score:1)
I'm a feeble-minded you insensitive clod!
Re: (Score:2)
I ask for it. Just to end this stupid "nobody asked for it" argument.
I also think a few million ChatGPT users asked for it. But maybe ask them yourself.
Re: How is this even "tech" anymore? (Score:2)
You and your few million fellow mouth-breathers deserve lobotomies.
Re: (Score:2)
I think the correct statement is "nobody smart asked for it". As most people are not smart, that checks out perfectly well.
Re: (Score:2)
You are probably using the "Free" or "Basic" versions that are basically like a worse google. The expensive paid versions can do actually useful stuff, it's only a matter of time (maybe years) before that usability rolls down to free or ad-supported versions
Re: (Score:1)
I sprang for the $200 ChatGPT a few days ago. It turns out not to be better than the $20 one. This could be related to a change that happened some time in the last week that seemed to make the model lazier.
Re: (Score:2)
It allows you to generate two pictures at a time instead of one. Well worth it. /s
(I use the $200 model, for a number of reasons, and I wish it could give more memory and longer chats before petering out.)
Re: (Score:2)
I know several people that tried the more expensive tiers up to $200. Mostly they found more complex hallucinations and the ability to give bad answers on more complex questions.
Incidentally, as still none of the AI pushers turns a profit, it is in no way certain this will ever "roll down" to the free tiers. In fact, it may go away completely because it is too expensive to run.
Re: (Score:2)
Incidentially, I am wondering if "free models" from China is a deliberate strategy to undermine the mindshare and apparent dominance of the all American OpenAI, Gemini, Microsoft thingie, etc. Why do we buy toothpaste at Walmart? Because it's the same toothpaste as at your corner store, just cheaper.
It would suggest a coordinated strategy, but coming from China, that would not be out of the question.
Re: (Score:2)
Well, good question. I guess it runs on tech and people are trying to get-rich-quick with it.
Re: (Score:2)
If it were not used for easily created media, disrupting communications, and being touted as a job replacer, we likely would have a different thing to say about AI in general. AI has also been the target and worry about people since the 50s. Especially when Hollywood has made it one of the pantheon of generic enemies (Terminator, 2001, etc.)
The main issue is how it is used. AI used for protein folding to create useful medicines, few would have issues with. AI used for botting, and psy-ops? Different st
But they are largely worthless (Score:2)
I've used those Chinese models at times for coding assistance and they are definitely inferior to the top proprietary models.
And then where does the money come from for open source models? Yes they are cheap or free to use, but if you actually want to get some quality work done you will pay a nickel or so per prompt. Even the top commercial models with hundreds of millions of weekly users aren't economically sustainable at present.
I've got no problem with the democratization of AI, and open source has been
Re: (Score:2)
I have used the Chinese models. Many can't draw, and the ones that I use for coding are not as versed as ChatGPT, Gemini, or CoPilot. Part of what makes a model useful is what data it is trained on, so it can give accurate items. With coding, having a model that has a lot of training on it helps greatly. Otherwise, one may wind up spending more time correcting the "vibe", than if one just wrote everything from scratch.
Some serious asterisks required. (Score:2)
> Chinese firms like Alibaba receive higher ratings than OpenAI and Meta on LMArena.
[1]No, they don't. [lmarena.ai]
Well, Meta- ya, because Meta is trash.
The one real world exception is the "CoPilot" category.
This category is problematic for several reasons.
It's missing the SOTA Google and OpenAI models, being the biggest.
Second, like 90% of its data is from Python, and the Chinese firms have Python-specific fine-tunes.
[1] https://lmarena.ai/leaderboard/
Re: (Score:3)
They are talking about downloads, not benchmarks. There are many benchmarks and many are pretty worthless as some models are overfit to the benchmarks and fail completely on tasks not in the benchmark.
Re: (Score:2)
They're not.
LMArena does not "rate" downloads.
I'll quote more of the summary for additional context:
> DeepSeek leads Hugging Face in popularity. Chinese firms like Alibaba receive higher ratings than OpenAI and Meta on LMArena. The site uses blind tests to measure user preferences.
1) Benchmarks are not useless merely because they're overfit. Overfitness for the benchmark still generally correlates with an increase in fitness for the real-world use scenarios.
2) LMArena isn't a benchmark. It's a blind user-preference statistics collector.
Re: (Score:2)
AI "scores" are crap. Otherwise anybody could immediately see how pathetic these things are.
Re: (Score:2)
Old_Man_Yells_At_Cloud.gif
Re: (Score:2)
Pathetic. And blind to reality. Not that this makes you special in any way.
Re: (Score:2)
Blind to reality- lol... Says the 21st century flat-earther.