Anthropic CEO says spies are after $100M AI secrets in a 'few lines of code'
techcrunch.com
techcrunch.com
> computers have failed to liberate humanity, and instead have "distorted and simplified our view of the world around us."
[1] https://darioamodei.com/machines-of-loving-grace
[2] https://en.m.wikipedia.org/wiki/All_Watched_Over_by_Machines...
DeepSeek, and Alibaba (Qwen) are making their models open source including the papers explaining “their secrets”.
Also from https://www.chinatalk.media/p/anthropics-dario-amodei-on-ai-...
> Jordan Schneider: We’ve talked about competition from the lens of compute, from the lens of models, but models are created by people. The times I’ve been to NeurIPS, it has been really striking to me just how much Mandarin is spoken at those poster sessions. I’m curious what message you’d have to say to PRC nationals studying in the West who hear this and are like, “Why would I want to work for this guy?"
> Dario Amodei: I want to be really clear on one thing — I should probably say it even more than I do. When we talk about China, this isn’t about Chinese people versus American people.
But we'll totally suspect you of being a spy for the PRC
I wonder how many western labs have not hesitated to implement techniques revealed openly by DeepSeek.
DeepSeek has open-sourced arguably more than 100M in value so far, from detailing their inference stack to discussing how they bypassed CUDA to use PTX[0] directly, or even open-sourcing an entire filesystem[1].
If it's revealed openly the it's not stealing. What's your point?
What deepseek has done is so much better than anything else in the West that the panic hasn't left yet.