We tried out DeepSeek. It worked well, until we asked it about Tiananmen Square and Taiwan.

[ 28/01/2025 07:07 ] Read on theguardian.com

Report Overview: The article presents a comparison of AI chatbots, highlighting differences in responses to politically sensitive topics. While it provides a balanced view of the chatbots' capabilities, it includes manipulations such as biased language, omission of key information, and selective use of sources, particularly in the portrayal of DeepSeek's responses.

Sides Objectivity Scores

Sides Representation Balance

Favored Side

ChatGPT and Gemini

Caution! Due to inherent human biases, it may seem that reports on articles aligning with our views are crafted by opponents. Conversely, reports about articles that contradict our beliefs might seem to be authored by allies. However, such perceptions are likely to be incorrect. These impressions can be caused by the fact that in both scenarios, articles are subjected to critical evaluation. This report is the product of an AI model that is significantly less biased than human analyses and has been explicitly instructed to strictly maintain 100% neutrality.

Nevertheless, HonestyMeter is in the experimental stage and is continuously improving through user feedback. If the report seems inaccurate, we encourage you to submit feedback , helping us enhance the accuracy and reliability of HonestyMeter and contributing to media transparency.

Detected Manipulations & Suggested Changes

Biased language

Use of language that unfairly favors one side over another.

The article uses phrases like 'censorship and suppression' and 'doomed to fail' when describing DeepSeek's responses, which could imply a negative bias.

Use neutral language when describing DeepSeek's responses, such as 'DeepSeek adheres to China's official stance on Taiwan.'

Omission of key information

Leaving out important details that could provide a fuller understanding of the topic.

The article does not mention the reasons behind DeepSeek's censorship, such as legal requirements in China, which could provide context for its responses.

Include information about China's legal requirements for AI content to provide context for DeepSeek's responses.

Selective use of sources

Using only certain sources or examples that support a particular viewpoint.

The article highlights instances where DeepSeek fails to provide information but does not equally emphasize where it performs well.

Include examples where DeepSeek provides accurate and useful information to balance the portrayal.

Spread the Truth

Share Report!

- This is an EXPERIMENTAL DEMO version that is not intended to be used for any other purpose than to showcase the technology's potential. We are in the process of developing more sophisticated algorithms to significantly enhance the reliability and consistency of evaluations. Nevertheless, even in its current state, HonestyMeter frequently offers valuable insights that are challenging for humans to detect.