I had a similar experience.

It looks like the LLMs weren’t trained for medical tasks. The study would be more interesting if it had been run on something built for the task.

tl;dr there’s a bunch of base64-encoded PDFs in the Epstein emails. Buddy had a hard time extracting them because the files are exported as images and the one and ell characters are almost indistinguishable.

Dude eventually managed to distinguish them. 🤷‍♂️

Huh. I wonder how it compares to codeberg.

Well, the shareholders want money. So it could solve that problem.