Artificial Intelligence thread
- Thread starter 9dashline
- Start date
thats deepseek v4 lite, so the full version is probably at parity with opus 4.6
here is the current size of China's AI data center build out.
one thing of note from today is that the Qwen team lead Junyang got pushed out.
A bunch of key Qwen personnel either quitting (or being fired). Either way, I've long been very harsh on Qwen which I viewed as mostly benchmaxxed garbage. Qwen 3.5 was extremely underwhelming. Alibaba has loads of cash so I don't think they will fold, but they clearly need a more talentdense team. Moonshot/Kimi has like 350 employees and yet it has a better model with a tiny budget.
Weight by weight they released the best open model ever that you can run locally. Nothing come close, most models close and open feel smart because they are just bigger. Well what comes around goes around, I hope Alibaba AI knows what is doing.
The Qwen team put China LLMs in the map way before Deepseek and other Chinese AI lab did. Hopefully they will land in an AI lab that allows them to continue their work and no be poached by a company who don´t care or worse some of these greedy, soulless US AI companies because that would be bad.
What Qwen haters don't understand is that Qwen models are basically the Unitree of LLMs, they are the standard in Research and Industry because their intelligence for size ratio is pretty good and that allowed industry and academia to finetune them better than other models.
US models are a close greedy mess and other Chinese models are just too big to be use in research.
They are not the "Unitree of LLMs". you are comparing apples to oranges here. Qwen is the de facto small model that all the embodied AI use. Regardless of how good the small models are, their large model just isn't up to par right now. I cannot possibly use Qwen max after using Kimi 2.5 or Minimax 2.5.
There are reasons why Alibaba management would be looking for a change.
Unitree humanoid robots are the most used humanoids in research bar to none and Qwen LLMs are the most used LLM in research. Kimi and Minimax are really good for end users but they are not research friendly.
As you said, Alibaba is looking to joint other Chinese labs in the commercial race as they should. Hopefully they will continue to open sourcing models because I believe is China AI strength. Diffusion of their models globally. despite efforts from the US.
But if Qwen goes down would be a lost in my opinion.