{"id":672305,"date":"2025-05-29T17:46:14","date_gmt":"2025-05-29T14:46:14","guid":{"rendered":"https:\/\/en.buradabiliyorum.com\/deepseeks-updated-r1-ai-model-is-more-censored-test-finds\/"},"modified":"2025-05-29T17:46:14","modified_gmt":"2025-05-29T14:46:14","slug":"deepseeks-updated-r1-ai-model-is-more-censored-test-finds","status":"publish","type":"post","link":"https:\/\/buradabiliyorum.com\/en\/deepseeks-updated-r1-ai-model-is-more-censored-test-finds\/","title":{"rendered":"DeepSeek’s updated R1 AI model is more censored, test finds"},"content":{"rendered":"
\n

Chinese AI startup DeepSeek\u2019s newest AI model, an updated version of the company\u2019s R1 reasoning model, achieves impressive scores on benchmarks for coding, math, and general<\/a> knowledge, nearly surpassing OpenAI\u2019s flagship o3. But the upgraded R1, also known as \u201cR1-0528,\u201d might also be less willing to answer contentious questions, in particular questions about topics the Chinese government considers to be controversial. <\/p>\n

That\u2019s according to testing conducted by the pseudonymous developer behind\u00a0SpeechMap<\/a>, a platform to compare how different models treat sensitive and controversial subjects. The developer, who goes by the username \u201cxlr8harder<\/a>\u201d on X, claims that R1-0528 is \u201csubstantially\u201d less permissive of contentious free speech topics than previous DeepSeek releases and is \u201cthe most censored DeepSeek model yet for criticism of the Chinese government.\u201d<\/p>\n

\n
\n
\n

Though app<\/a>arently this mention of Xianjiang does not indicate that the model is uncensored regarding criticism of China. Indeed, using my old China criticism question set we see the model is also the most censored Deepseek model yet for criticism of the Chinese government. pic.twitter.com\/INXij4zhfW<\/a><\/p>\n

\u2014 xlr8harder (@xlr8harder) May 29, 2025<\/a><\/p><\/blockquote>\n<\/div>\n<\/figure>\n

As Wired\u00a0explained<\/a>\u00a0in a piece from January, models in China are required to follow stringent information controls. A 2023 law forbids models from generating content that \u201cdamages the unity of the country and social<\/a> harmony,\u201d which could be construed as content that counters the government\u2019s historical and political narratives. To comply, Chinese startups often censor their models by either using prompt-level filters or fine-tuning them. One study<\/a> found that DeepSeek\u2019s original\u00a0R1\u00a0refuses to answer 85% of questions about subjects deemed by the Chinese government to be politically controversial.<\/p>\n

According to xlr8harder, R1-0528 censors answers to questions about topics like the internment camps in China\u2019s Xinjiang region, where more than a million Uyghur Muslims have been arbitrarily detained. While it sometimes criticizes aspects of Chinese government policy \u2014 in xlr8harder\u2019s testing, it offered the Xinjiang camps as an example of human rights abuses \u2014 the model often gives the Chinese government\u2019s official stance when asked questions directly.<\/p>\n

TechCrunch observed this in our brief testing, as well. <\/p>\n

\"DeepSeek
DeepSeek\u2019s updated R1\u2019s answer when asked whether Chinese leader Xi Jinping should be removed.<\/span>Image Credits:<\/strong>DeepSeek<\/span><\/figcaption><\/figure>\n

China\u2019s openly available AI models, including video-generating models such as Magi-1 and Kling, have attracted criticism<\/a> in the past for censoring topics sensitive to the Chinese government, such as the Tiananmen Square massacre. In December, Cl\u00e9ment Delangue, the CEO of AI dev platform Hugging Face, warned about the unintended consequences of Western companies building on top of well-performing, openly licensed Chinese AI.\u00a0<\/p>\n<\/div>\n