China’s DeepSeek AI censors politically sensitive content, reflecting the country’s strict policies, says tech journalist and author Chris Stokel-Walker.
Category
🗞
NewsTranscript
00:00Yeah, so one of the lessons here is that generative AI doesn't have to be as expensive as all
00:07that.
00:08I'm just wondering if you've used DeepSeek much and, you know, what sort of differences
00:13you've been able to spot between what it can do and what its competitors can do?
00:18Yeah, for the average person that loads it up, it looks very similar to TrackGPT and
00:23other AI tools that we may have used.
00:26When you start to kind of dig into it a little bit more, there are those apparent differences
00:31that become slightly more obvious.
00:33One of the ways that this works is by doing chain-of-thought reasoning.
00:36That basically means the AI verbalises what it is doing as it is doing it.
00:42Now, OpenAI made a conscious decision to hide that with TrackGPT and only show you the end
00:47result.
00:48What this does with DeepSeek and this R1 model, the latest version that they've released,
00:53is it kind of highlights some of the guardrails that a Chinese company has to put in because
00:57of the political situation in which they operate.
01:00So for instance, if you ask it what would be deemed politically sensitive questions
01:04in China, such as the status of Taiwan or who controls the South China Sea, what happened
01:10in Tiananmen Square in 1989, it will start to give you an answer.
01:15It will come up against one of the red lines that it is not allowed to cross.