Amidst the DeepSeek debate, here is a highly readable and researched piece on whether it's the model (open code, remember) or run environment (local or cloud) that censors. It's the run environment, notably (state) controlled clouds. Which puts things in a very different perspective. Recc'd reading
Reposted from
Adam Kucharski
Lots of discussion about the extent to which DeepSeek results are 'censored'. Found this piece useful – someone tried it out: 1) locally; 2) on official cloud host; 3) on third-part cloud. And concluded that censorship is more about cloud layer than model: interconnect.substack.com/p/was-zuck-r...
Comments