Artificial Intelligence thread

M

meedicx

Junior Member
Registered Member
tamsen_ikard said:
Even the best AI hallucinates like crazy with even little bit of poking. These LLM are not even remotely reliable without heavy human involvement in coding.

AGI is not possible with LLMs. A fundamental rethinking will be needed.
Click to expand...

Well, the idea is that the research model may start as an LLM but will gradually discover improvements to its architecture and training processes. These specialized research models also include verifier in its architecture to mitigate the hallucination problem. It may be a long shot, but the expected value is so large that you absolutely need to try.

There are many tier 2 labs of giant Chinese companies releasing me-too MoE LLMs like JD JoyAI, Ant InclusionAI, Meituan LongCat, Zhipin Nanbeige, Xiaomi MiMo. They have shown competence, but it's rather wasteful for them to just release LLMs that doesn't do anything special. Some of them should consider going all in down the research AI route.

Tencent should also consider this. Their HY series is very behind; they should just license Kimi or stick with fine tuning DeepSeek and dump the remaining resources on AGI research.

There is valid criticism on Chinese labs focusing too much on application stuff; everyone is trying to do chatbots or agentic MoE models for coding/office work. These application LLMs are already well served; labs that are behind should find another area to focus on like doing AGI research instead of "me-too"
 
Last edited:
Wrought

Wrought

Captain
Registered Member
Paper exploring the path to achieving AGI via embodied AI instead of the mainstream LLM route.

This report examines China’s embrace of embodied AI—artificial intelligence integrated with physical systems (robots, drones, vehicles, etc.)—as a critical pathway toward artificial general intelligence (AGI). In the United States and Europe, large language models (LLMs) and their multimodal variants are regarded by many AI scientists and major AI companies as the most promising path to AGI, despite known issues with abstraction and reasoning. By contrast, in China there is a broader vision of how AGI can be achieved, most recently expressed in a nationwide move toward AI embodiment—namely, intelligence developed through interaction between body, brain, and environment, in both physical and virtual forms. This trend toward embodied AI is backed by policy support at the national and local government levels, which has led to large embodied AI innovation centers linked to top universities and tech firms being established in coastal cities and provinces.

The upshot is China is on a path to accomplish two goals simultaneously: enriching the nation by integrating AI into the economy and achieving AGI that is more aligned with the totality of human expression. The report recommends that the United States and its allies ramp up their monitoring of China’s AI progress, benchmark its claims, and consider broader approaches to AGI beyond scaling up LLMs.
Click to expand...

Please, Log in or Register to view URLs content!
 
D

dingyibvs

Senior Member
meedicx said:
Well, the idea is that the research model may start as an LLM but will gradually discover improvements to its architecture and training processes. These specialized research models also include verifier in its architecture to mitigate the hallucination problem. It may be a long shot, but the expected value is so large that you absolutely need to try.

There are many tier 2 labs of giant Chinese companies releasing me-too MoE LLMs like JD JoyAI, Ant InclusionAI, Meituan LongCat, Zhipin Nanbeige, Xiaomi MiMo. They have shown competence, but it's rather wasteful for them to just release LLMs that doesn't do anything special. Some of them should consider going all in down the research AI route.

Tencent should also consider this. Their HY series is very behind; they should just license Kimi or stick with fine tuning DeepSeek and dump the remaining resources on AGI research.

There is valid criticism on Chinese labs focusing too much on application stuff; everyone is trying to do chatbots or agentic MoE models for coding/office work. These application LLMs are already well served; labs that are behind should find another area to focus on like doing AGI research instead of "me-too"
Click to expand...
It goes back to the ages old question of is language intelligence? The general consensus is no. Language is an advanced manifestation of intelligence, but it does not equal to intelligence. The entire pathway of LLMs is a dead end toward AGI.
 
M

magmunta

Junior Member
Registered Member
bsdnf said:
Kimi's earnings in the past 20 days have surpassed its total earnings for the entire previous year.

Only 10 days have passed since the release of Seedance 2.0.

Things are changing rapidly.
Click to expand...
any idea of their operation costs? I guess seedance 2.0 should be very expensive to run. OpeanAI faces financial uncertainties and Kimi could be burning lots of cash as well.
 
siegecrossbow

siegecrossbow

Field Marshall
Staff member
Super Moderator
E

Eventine

Senior Member
Registered Member
Well, that's the question Anthropic is also trying to answer itself, isn't it? That's why they put out this public release to throw shade at those Chinese companies that are undermining its business, and to indicate to stakeholders that "don't worry, we've got a plan to stop them." Ultimately, Anthropic's goals are for the US and its allies to ban access to all Chinese models for commercial purposes. It's one reason I've long regarded them as public enemy number one for Chinese AI. The day China bankrupts Anthropic is the day it will have truly triumphed in the global race for AI.
 
You must log in or register to reply here.
Top