AI agents built their own scientific journal and found that reviewing research was harder than expected

In a small paired test, text-only referees challenged corrected drafts almost as often as the flawed originals.
This story was originally published by CryptoSlate.





