Artificial Intelligence thread
- Thread starter 9dashline
- Start date
Im disappointed to see what US AI are driven to when they face outside competition.
Platform dominance is and has been the main strategy by which American Big Tech maintains their trillion+ dollar valuations and outrageous profits & influence on the world. China is by far their biggest obstacle to achieving this same result in AI, and the only hope for breaking the Western monopoly in software platforms.
This is pretty much why the US and American companies have no chance against China in technological war. Where American companies gain by establishing and maintaining effective monopolies that block out smaller and newer operations, China builds its companies through brutal competition. While Americans constantly preach the virtues of capitalism, I don't think that they practice real capitalism any more. Instead, they've latched onto corporatism and plutocracy just like that found in any cyberpunk story that's worth reading/watching/playing. This is also why nobody really cares if any Chinese company suffers some sort of major setback. If something like Deepseek were to fail, there will be a hundred companies looking to take its place, each hungrier than the last. I can't say the same if this were to happen to any American company of note.
Another thing I've noticed is that American AI companies seem to promise a lot of things and deliver theoretical advances, but it's the Chinese companies that seem able to put AI into practise and mange to deliver real-world benefits. We are already seeing AI integration into all manner of Chinese industry and everyday life, whereas it's still mostly a toy in the US.
CSCEC with new data center in Xinjiang of 20 EFLOPS. Has a mix of ascend-910C-2 chips, RTX-4090 and H200
A lot of news in the US as of late with respect to the AI talent wars, how companies are now publicizing their latest recruitment of top tier engineers. I could only imagine what kind of fees the recruiters who facilitated their onboarding are getting right now, and to think all they had to do was be on a higher tier of surfing Linkedin. (I've always hated that, I'm of the opinion even Amazon drivers deserve to make more money than they do) More to the point, half of the names listed in recent days from both Meta and Nvidia are Chinese. No matter where they be, ethnic Chinese are still the cream of the crop when it comes to AI talent.
Baidu open sourced a whole bunch of Erniebot models.
Zhipu's new MaaS: VLM with the best 10B effect is launched, and Agents application space is launched
The Zhipu Open Platform Industry Ecosystem Conference was held at the Zhangjiang Science Hall in Pudong, Shanghai. The conference brought together government leaders, top developers and corporate users to present Zhipu's core achievements in multimodal intelligence and MaaS ecology. At the conference, Pudong Venture Capital Group and Zhangjiang Group jointly announced a strategic investment in Zhipu, injecting solid momentum into Zhipu's construction of a trusted artificial intelligence infrastructure.
In his keynote speech, Zhang Peng, CEO of Zhipu, announced two latest achievements of Zhipu and its ecological partners in moving towards AGI: one is the open source release of the new generation general visual language model GLM-4.1V-Thinking , which takes reasoning ability as the core breakthrough and refreshes the performance ceiling of the 10B-level multimodal model; the other is the launch of the new MaaS Agent aggregation platform "Application Space" , which fully activates AI capabilities in industry scenarios and links with Z Fund to launch a special support plan for Agent pioneers.
At the same time, in the keynote speech session of the conference, Wu Weijie, senior vice president of Zhipu, Li Chengjie, vice president and chief digital intelligence officer of Mengniu Group, and Lv Xufeng, deputy director of China UnionPay Financial Technology Research Institute, respectively shared their views on the implementation of big models. In the panel session, Hu Xiuhan, founder of NieTA, Wang Zhentong, co-founder of AiPPT, Guaizi, CMO of Flowith, and Shen Ling, general manager of Zhangjiang Zhihui, exchanged views on the topic of bigmodel native entrepreneurship.
he model performs particularly well in the following tasks, demonstrating high versatility and robustness:
- Image General: Accurately identify and comprehensively analyze image and text information;
- Math & Science: Supports complex problem solving, multi-step deduction, and formula understanding;
- Video Understanding (Video): Capable of time series analysis and event logic modeling;
- GUI and web agent tasks (UI2Code, Agent): understand the interface structure and assist in automated operations;
- Visual anchoring and entity positioning (Grounding): Accurately align language and image areas to improve the controllability of human-computer interaction.
Currently, GLM-4.1V-9B-Thinking has been open-sourced simultaneously in Hugging Face and the MoDa community. It contains two models: GLM-4.1V-9B-Base , which is expected to help more researchers explore the capabilities of visual language models; and GLM-4.1V-9B-Thinking , which is a model with deep thinking and reasoning capabilities. This model is used for normal use and experience.