That's what I've been saying as well. We don't even know how to define or quantify human intelligence, how can we build something that surpasses it? We don't even know what we're shooting for.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
That's what I've been saying as well. We don't even know how to define or quantify human intelligence, how can we build something that surpasses it? We don't even know what we're shooting for.
yes, both Alibaba's Z-Image and DeepSeek's Math V2 are big stories.
Z-Image already had half a million download and is the highest trending model on HF.
6B parameters, realistic image quality (especially the lighting), accurate bilingual text rendering, powerful command compliance, and NSFW supported—all of these can run locally with only 16GB of VRAM.
And this is just the distilled Turbo version; the full Base and Edit version will destroy a large number of orders for closed-source commercial APIs.
Last edited:
Furthermore, Whisper Thunder, a SOTA-level video generation model, is currently in open testing in the arena and is ranked number one. We don't know which company it's from; some speculate it's a Chinese model.
Interestingly, I once saw a comment on Bilibili from someone claiming to be an employee of Google's Nano Banana team. He stated that it was just a silly, joking codename, they didn't expect the name to generate such anticipation for the model (of course, also related to the model's quality). And now everyone just naming their unreleased model
Interestingly, I once saw a comment on Bilibili from someone claiming to be an employee of Google's Nano Banana team. He stated that it was just a silly, joking codename, they didn't expect the name to generate such anticipation for the model (of course, also related to the model's quality). And now everyone just naming their unreleased model
Last edited:
broadsword
Brigadier
A few months ago, I uploaded my full blood test report to Google AI, ChatGPT and Deepseek for their diagnosis. My haemoglobin was slightly below the recommended range.
Google AI was negative and alarmed, and advised me to consult a doctor.
ChatGPT was more neutral, but advised to consult a doctor as well.
Deepseek was neither. Instead, it determined that I was a-ok after taking into account the other factors, which was the same conclusion as what my doctor arrived at.
So all those tests where it trailed the Western models puzzled me.