Re:Interesting Interview with DeepSeek's CEO
by ramblenode on Tuesday December 31, 2024 @08:50PM
(#42563346)
> Chinese models have to censor a long list of words that threatens the government, which makes them super dumb.
The heavy-handed curation and self-censorship of ChatGPT and Gemini responses is literally a meme, though. Or are you referring to the training data?
2.) Chinese models have to censor a long list of words that threatens the government, which makes them super dumb. List of stupid words example: sprinkle pepper, accelerationism, my emperor, lifelong control, etc. and the list of censored words grow(!!) as Chinese citizens try different combination of words to escape censorship.
3.) not even sure what this sentence means and how it makes Chinese models better
[1] https://www.bloomberg.com/news/articles/2024-12-09/china-is-...