Artificial Intelligence thread

tphuang

tphuang

General
Staff member
Super Moderator
VIP Professional
Registered Member
Can we please get back on topic with AI?


Please, Log in or Register to view URLs content!

Here is an article from FT on the wide adoption of DeepSeek in China.

“The pace at which DeepSeek is spreading is incredible. Before, conservative institutions like government agencies and hospitals were nervous about introducing generative AI applications, fearing trouble if something goes wrong,” said one Hangzhou-based AI engineer.
Click to expand...
Local governments, including Jinan’s and Hangzhou’s, have launched citizen chatbots built on DeepSeek so that residents can ask questions about everything from tax payments and rubbish collection to birth certificates. The city of Shenzhen’s Futian district has launched multiple AI agents built on DeepSeek models, including a document generation tool for law enforcement officers to draft administrative reports.
Click to expand...
There is definitely a mentality change here.

HSBC tech hardware analyst Frank He wrote in an analyst note that DeepSeek “AI inferencing workloads have been surging, triggered by the growing popularity of DeepSeek R1 in recent weeks”. He predicted demand would continue and “trigger related hardware and software upgrades in cloud computing and AI infrastructure”.
Click to expand...

Even the quants have now shifted to all using AI for their trading

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!
According to the
Please, Log in or Register to view URLs content!
, the AI startup is currently being used by at least 10 state-owned energy companies and 13 city governments.
Click to expand...
I've personally lost count, but SOE and regional government adoption is happening at lightning speed really (for them)

Please, Log in or Register to view URLs content!
Looks like they are now reporting jump in demand for H20s from Tencent, ByteDance & Alibaba. I'm not surprised here. It's kind of necessary, but they are also procuring domestic chips in big numbers. They are snapping up anything they can.
 
A

AndrewS

Brigadier
Registered Member
Wrought said:
Yes that's right when it comes to real-world decisions, but your original assertion was about economic theoreticals, and it was wrong.
Click to expand...

Uh no.

Eventine posted a long spiel about the theory of comparative advantage relating to AI.

My response pointed out how it doesn't work in the example of US and China.
 
E

Eventine

Junior Member
Registered Member
mossen said:
Kimi would like to remind you that they are not dead.

View attachment 146582


Their latest public release is 1.5. So they are clearly going to release something big soon.
Click to expand...
Without an open source / open weights release, they're going to have to show more if they want to be noticed. Matching o1 performance is, by now, not particularly exciting given the half dozen other companies that's done it. Grok 3 has the uncensored angle going for it, Deep Seek is open weights, Anthropic is still the king of practical coding, and Google has by far the cheapest API (at ~1500 free requests per day). Even Qwen is struggling for attention because it's just not doing anything special any more.

Eyes right now are on the next generation of models - potentially led by ChatGPT 4.5 and o3 - and Deep Seek v4 / R2. Meta may also throw a curve ball there if they manage to get LLama 4 out this year and reach state of the art.

The field is moving quickly and AI companies need differentiation to help them get noticed and survive.
 
F

Fatty

Junior Member
Registered Member
Please, Log in or Register to view URLs content!

GPT 4.5 released. By OpenAI’s own metrics, only slightly better than GPT 4o. OpenAI also says it is not a frontier model, but is their largest model yet with 10x better efficiency. Honestly seems like a very mediocre release, especially given that it’s behind the 200/mo subscription. Personally, I’m disappointed in Altman’s hyping of it as “AGI.”

Also, now that people have gotten their hands on Claude Sonnet 3.7, there are some rumblings about it being worse than 3.5 still.
 
P

PantsFullOfAnts

Just Hatched
Registered Member
GPT 4.5, Grok 3, and Sonnet 3.7 have shown that pretraining has hit a wall.
I suspect the next big breakthrough will involve parallel chains of thought, which DeepSeek should be able to do better than anyone.
With Kimi 1.6 and other recent models, it seems that China is now in the lead with LLMs, and I don't expect this to change in the future.
 
Wrought

Wrought

Senior Member
Registered Member
AndrewS said:
Uh no.

Eventine posted a long spiel about the theory of comparative advantage relating to AI.

My response pointed out how it doesn't work in the example of US and China.
Click to expand...

No, @Eventine specifically pointed out how politics had overwritten comparative advantage in this context. And he was right to do so.

Eventine said:
But as we know, the situation wasn't allowed to develop organically. The US made trade with China a national security issue, and tried to force a decoupling. At the same time, China also banned US services from the Chinese ecosystem. The result is that comparative advantage was overwritten by politics, and now Chinese AI companies do have incentives to serve both domestic manufacturing industries and domestic service industries.
Click to expand...

Your response, that comparative advantage doesn't apply in the case of absolute advantage, was wrong. It absolutely applies, but in this case politics was the deciding factor, not economics.
 
You must log in or register to reply here.
Top