The Zero model is pure RL, no SFT, no cot traces... its trained on nothing, literally
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
And ChatGPT thinks in Chinese, lol
But here's the thing, the fact that Deepseek is MIT licensed, which allows commercial use, the fact that it's orders of magnitudes more efficient (read profitable), and the fact that closed-AI companies don't have to disclose their own code, means there is 100% chance all US closed-AI are directly using Deepseek under the hood.
Enuff said.
Look at them shamelessly groveling and reverse engineering a Chinese effort.
Deepseek R1 already being compared to Wuhan Lab Leak and Cyhna virus lol
Surprised it hasnt been classifed as weapons grade tech under ITAR
Hugging Face's smallest AI visual language model is launched: 256 million parameters, even PCs with less than 1GB of memory can handle it
The Hugging Face platform published a blog post yesterday (January 23), announcing the launch of two lightweight AI models, SmolVLM-256M-Instruct and SmolVLM-500M-Instruct, to maximize the computing power performance of devices with limited AI computing power.
The Hugging Face platform published a blog post yesterday (January 23), announcing the launch of two lightweight AI models, SmolVLM-256M-Instruct and SmolVLM-500M-Instruct, to maximize the computing power performance of devices with limited AI computing power.
So Deepseek kills over 1 million American companies and over 5 million Indian companies? Oorah!
It's the company own employees that leaked it
AssassinsMace
Lieutenant General
Yes China just blew the lid off the US scam of getting large amounts of subsidies in the billions… trillions funneled to US tech companies to produce AI. I think it was the idea where the US was eventually to have their AI go open source meaning get as much money as soon as possible before it does. That’s why Biden targeted against China getting AI chips to stop China from closing down that money train. What happens as usual? China foils their dastardly plot. China did for them earlier what they said they were planning to do.
The “forced” release of Operator/Agent ahead of schedule…?
The “forced” release of Operator/Agent ahead of schedule…?
Last edited:
DEEPSEEK WAS ABLE TO BUILD THIS IN A CAVE!!!
WITH A BOX OF SCRAPS!!!
Baichuan Intelligence releases deep thinking model Baichuan-M1-preview, which has reasoning capabilities in three major areas: language, vision, and search
In terms of visual reasoning capabilities, its performance in authoritative evaluations such as MMMU-val and MathVista has surpassed models such as GPT-4o, Claude3.5 Sonnet, and QVQ-72B-Preview.
In terms of visual reasoning capabilities, its performance in authoritative evaluations such as MMMU-val and MathVista has surpassed models such as GPT-4o, Claude3.5 Sonnet, and QVQ-72B-Preview.