As far I understand apart from the synthetic data I think DeepSeek bought their training data, so is probably more bias towards English and Chinese. But the model is open source and can be finetuned using Russian data relatively easy.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
This is crazy growth rate, fastest adoption in the world by far....
What an interesting phenomenon. I hope the explanation is not so crass as OpenAI just copy-pasting R1's open-sourced weights.
Now that Huawei Ascend is inferencing for DeepSeek and as the cost of intelligence trends towards electricity price to run the inferencing...
China needs to give AI to the world for free, not just in terms of open source and open weights but actual inferencing as well... chat app, web browser, api
(or at the very least, make SOTA models available priced at cost)
In the words of Steven Bannon, killing OpenAI is 100x more important....
China must deprive USA from weaponising AI and their evil plans of wholesale conversion of value of human intellect and labor worldwide....
Sam Altman made clear his intent to be God and enslave all the rest of humanity whom arent "chosen"
Their rationale is that if o3 etc can save a company the need for a PhD, then they feel that is what it should be priced at, even if the actual cost for them to run inference is 100x cheaper
For every $2 China spends to subsidize AI for the world, thats $58 the US via OpenAI isnt making....
China has subsidized the world before, it needs to do so again... this AI race is 1000x more important than tiktok, temu, solar, 5g, or EV
in Scaman's mind, you are cheating him out of the $499.50 that he already feels entitled to....
no wonder "Open"AI is lobbying hard to make open source illegal
Last edited:
Really intriguing that Llama didn't spark the same development compared to DeepSeek
Baidu Smart Cloud Qianfan fully supports DeepSeek-R1/V3 calls
"Baidu Smart Cloud" announced today that the DeepSeek-R1 and DeepSeek-V3 models have been officially launched on the Baidu Smart Cloud Qianfan platform.
The model connected this time has fully integrated the Qianfan reasoning link and Baidu's exclusive content security operator to achieve model security enhancement and enterprise-level high availability guarantee. It also supports comprehensive BLS log analysis and BCM alarms, helping users to build intelligent applications safely and stably.
According to reports, the Qianfan platform is committed to providing users with full-process, one-stop AI services. In addition to powerful model resources, it also matches a complete one-stop model effect tuning tool chain, including data processing, model fine-tuning, model evaluation, model quantification and other key links, to help companies deeply optimize model performance according to their own business needs. At the same time, the Qianfan platform has excellent model reasoning hosting capabilities, supports various mainstream reasoning frameworks such as vLLM, LMDeploy, TensorRT-LLM, SGLang, and also supports custom import and deployment of models, providing developers with a highly flexible development environment.
It did spark the same development, Llama was the one who sparked the open LLM revolution and that include DeepSeek.
Yet another try at DeepSeek's GPU number
This Korean guy usually post legit stuff on NAND and DRAM news.
This Korean guy usually post legit stuff on NAND and DRAM news.
Last edited:
Qwen has moved up the Elo rating leaderboard. 3 out of top 10 are Chinese models now.