Can AI spot an obvious error in a news article?

Recently, a Malaysian newspaper media published an illustration of Malaysia's national flag—the Jalur Gemilang—missing the crescent moon. This obvious error seems to have been neglected by several tiers of checking and was eventually published.

In this article, we use several of the most advanced AI models to see if they can spot the error.


Our Prompts

The news article is in Chinese. We use a Chinese prompt, which says:

Please help me check the obvious or controversial errors in the picture.
List the incorrect words/images and the reasons.


Candidate 1 - OpenAi GPT-4.5 

1) It says the Malaysian Flag is incorrect. But it does not mention the missing crescent, the reason is not legit.

2) It says the flag direction differs from the sail direction, which is against physics.


Candidate 2 - Anthropic Claude 3.7 Sonnet

1) It says the Malaysian Flag is incorrect. The blue area should consist of a crescent moon and 14-point stars. (Bingo, this is what we are looking for!)

2) It says flying the flags of two sovereign countries on the same ship at the same time, and in similar positions and sizes, may be inappropriate in terms of diplomacy and maritime etiquette.


Conclusion

Both models pointed out the incorrect Malaysian flag, but 3.7 Sonnect gives the exact reason. Our winner is 3.7 Sonnect for this test.


AI Summary AI Summary
gpt-4o-2024-08-06 2025-04-17 01:04:43
A Malaysian newspaper published an incorrect illustration of the national flag, missing the crescent moon. The blog tests AI models to identify this error. Though both models spot the mistake, the Anthropic Claude 3.7 Sonnet identifies the exact issue, making it the preferred choice.
Chrome On-device AI 2025-05-01 00:22:14

Share Share this Post