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


The title they went with How Open Must Language Models be to Enable Reliable Scientific Inference? Noisy translates that to

Closed AI models now fail scientific tests. Openness is the only fix.


Researchers have found that using AI models where details about their construction are hidden makes it hard to trust scientific results. To fix this, they say AI used in research must be more open about how it was built and used.
For years, scientists have been adopting AI tools for research. This paper points out a problem: if the AI is a 'black box,' its results can't be reliably verified. This means research using closed AI might be flawed. It suggests that for AI to be useful in science, its inner workings must be transparent, or at least fully documented. This could slow down the adoption of some AI tools but make scientific findings more trustworthy.
Watch whether research papers using AI now include detailed justifications for why a specific model was chosen and how its potential flaws were addressed.

If you insist
Read the original →