Noisy
← this week  ·  Subscribe →

The world is being quietly rearranged by people who write very long documents.

arXiv · April 10, 2026
The title they went with Beyond Surface Judgments: Human-Grounded Risk Evaluation of LLM-Generated Disinformation Noisy translates that to

AI tools cannot measure how disinformation affects humans, only how it affects other AI

What happened
Companies use AI tools to check if generated disinformation is dangerous. But it turns out these tools mostly agree with each other, not with actual human readers.
Why it matters
Companies use AI tools to decide if generated content is dangerous. This paper shows those tools are bad at predicting how real people react to disinformation. They might be filtering for the wrong things, or missing what actually works on human minds.
The signal
Watch whether social media platforms or content moderation services change their methods for evaluating AI-generated disinformation, or if they keep using AI judges.
If you insist
Read the original →