Artificial Intelligence thread

[mossen]

Okay, so we have the full results from Artificial Analysis. I looked up the hallucination benchmark and Qwen 3.7 Max is now proper SOTA.




Incredible job, Qwen team! Finally, this is a big Qwen model that I feel like I can use.

 

[siegecrossbow]

Please, Log in or Register to view URLs content!

“Man,” said the Ghost, “if man you be in heart, not adamant, forbear that wicked cant until you have discovered What the surplus is, and Where it is. Will you decide what men shall live, what men shall die? It may be, that in the sight of Heaven, you are more worthless and less fit to live than millions like this poor man’s child. Oh God! To hear the Insect on the leaf pronouncing on the too much life among his hungry brothers in the dust.

 

[Michael90]

But no mention of DeepSeek in its selected partners. They included everyone else

Interesting. Any idea why they omitted only Deepseek?

 

[bsdnf]

Gemini had previously been mocked as the ‘US Doubao’ for its repeated blunders; who would have thought that Flash 3.5, instead of shedding its derogatory reputation, it solidified that label within hours. Its knowledge base is even stuck in January 2025.

It might not even be as good as Doubao. Doubao may be clunky, but its multimodal capabilities and app experience are top-notch.

like what?

 

[9dashline]

Okay, so we have the full results from Artificial Analysis. I looked up the hallucination benchmark and Qwen 3.7 Max is now proper SOTA.

View attachment 175303


Incredible job, Qwen team! Finally, this is a big Qwen model that I feel like I can use.

I'm waiting for qwen3.7 9b... It's already smarter than most humans

 

[meedicx]

From the Qwen 3.7 max blog post, they described how the model optimized its own inference kernel achieving a 10x speedup on Alibaba's self-developed ZW-M890 PPU AI chips. This is another data point reinforcing my previous point that CUDA has no moat in the age of agentic coding. Ironically, the success of CUDA training AI coding models has destroyed its own moat.

Please, Log in or Register to view URLs content!

 

[tphuang]

https://twitter.com/i/web/status/2057430888573526380

Huge battle between Alibaba & ByteDance in AI has now found a new target in short videos as both players launch new platforms on that today.

 

[tphuang]

https://twitter.com/i/web/status/2057509502488875407

Interesting now to see CapCut features go into Gemini App. I don't know if this means Gemini will start using ByteDance's models.

 

[9dashline]

https://twitter.com/i/web/status/2057509502488875407

Interesting now to see CapCut features go into Gemini App. I don't know if this means Gemini will start using ByteDance's models.

Hmmm Google couldnt use Gemini to code a capcut themselves?

 

[tphuang]

Please, Log in or Register to view URLs content!

Huawei unveils new AI DC full stack solution with its own Agent framework, model engine/compute platform, memory storage, AI Data Ocean & such.