The Wall Street Journal is reporting that Baidu will soon integrate ChatGPT into the company’s chat/search offerings. The company plans, however, to:
limit its chatbot’s outputs in accordance with the state’s censorship rules, one of the people said. OpenAI also applies restrictions to ChatGPT’s outputs in an effort to avoid toxic hate speech and politically sensitive topics.
While I have no doubt that Baidu will impose censorship, I wonder whether researchers will be able to leverage the learning properties of ChatGPT to gain insight into what is censored by Baidu. Side-channel research has been used to reveal how censorship is undertaken by companies operating in China; I’d expect using these AI models will offer yet another way of interrogating their censorship engines.