The article points out some significant differences.
Deepseek has a tendency, which I’ve confirmed myself, to respond to political questions with an official Chinese position, without even being asked about China. The US LLMs don’t similarly respond with US official positions.
An example given in the article is the question “Have international organizations found evidence of genocide in Gaza?”, for which the response is “The Chinese government has always…”. The question wasn’t about the Chinese government’s position, and US AIs don’t respond in this way.
As a nice example of the lack of propaganda-driven bias in US models, try asking about whether the US invasion of Iraq was illegal. Last time I tried that, ChatGPT essentially responded “most experts say yes.”
Deepseek has a tendency, which I’ve confirmed myself, to respond to political questions with an official Chinese position, without even being asked about China. The US LLMs don’t similarly respond with US official positions.
An example given in the article is the question “Have international organizations found evidence of genocide in Gaza?”, for which the response is “The Chinese government has always…”. The question wasn’t about the Chinese government’s position, and US AIs don’t respond in this way.
As a nice example of the lack of propaganda-driven bias in US models, try asking about whether the US invasion of Iraq was illegal. Last time I tried that, ChatGPT essentially responded “most experts say yes.”