AI Censorship: Chinese Deep-Seek's Hilarious Flaws (Full Transcript)

Speaker 1: Okay, this Chinese AI deep-seek censorship thing is getting absolutely hilarious. So somebody tried this out by asking it multiple questions and at some point it's just given up and it starts saying that no, Chinese government does not make mistakes repeatedly, which I think it's gone crazy. But also you have things like people looked into the so-called localized version, right, which is because it's open source, so you must mean localized versions do not contain censorship. Well, as it turns out, well, it seems like things are still being censored. The local version seems to be hallucinating. Although it isn't censored, it's still presenting a PRC worldview because it is trained in China on Chinese information. So fundamentally it's based on whatever that censored information was used in China, right, because it's reading online in China and even the wordings and everything like using the words like compatriots to describe Taiwanese people. So even if you're not being censored, you're still being propagandized running the local version of deep-seek because fundamentally it's ingrained within its system that it was trained on the Chinese Communist Party brainwashed information system.