有没有人用过deepseek_v3? Zuckerberg 都怕了

[billwanhua]

1. 开源
2. 各个关键测试不差于美国各大AI
3. 最关键的是所用的参数(671 billion parameters, with 37 billion parameters actively engaged during inference)比美国各大AI 模型（chatgput4 is estimated to have 100 trillion parameters. )低一个数量级
4. 运行效率比比美国各大AI 模型高一个数量级

View: https://www.reddit.com/r/LocalLLaMA/comments/1i2y810/is_deepseek_v3_overhyped/


View: https://youtu.be/8EYKlXso718?si=VpwumvuHB6J275om

DeepSeek's new AI model appears to be one of the best 'open' challengers yet | TechCrunch

ChCC

techcrunch.com

免费的，大家上网试试吧：

https://chat.deepseek.com/

目前这个AI 模型应该是性价比最高的，真不能小瞧中国的科技进步, 那些笑话中国抄袭的应该笑自己

 

[billwanhua]

DeepSeek-R1 and exploring DeepSeek-R1-Distill-Llama-8B

DeepSeek are the Chinese AI lab who dropped the best currently available open weights LLM on Christmas day, DeepSeek v3. That model was trained in part using their unreleased R1 …

simonwillison.net

有lite model 可以用在笔记本电脑了，只需要CPU和48G内存，以前有人问Facebook Llama 能不能在家里电脑运行，好像不行

 

[向问天]

我早试过了。的确了不起。
拜登禁止AI芯片进入中国，完全禁了个寂寞。

 

[tt8966]

苹果店有deepseek 手机app