A new paper compares Gemini with Open evidence with ChatGPT: the winner is....

... It doesn't matter the paper is already obsolete

A new paper in Nature Medicine compares the accuracy of Gemini with Claude with OpenEvidence with Uptodate with ChatGPT. The main result is making a splash online. But ‘experts’ are missing the main point.

Forget Gemini, here is what people are missing:

These platforms were tested in the fall of 2025 and most recently February 2026.

The paper was published in June 2026.

It's 4 months (minimum) old.

The paper is already obsolete.

Here is the lesson:

Academic medicine is too slow to study these products. In fact, the entire do research, write it up, submit an article to a corrupt publisher (look up Nature’s profits), wait for editors and reviewers (who often do not know anything) to assess your paper, proof it, edit it, typeset it… it's too slow. The result is useless by the time it appears.

Elsewhere I explain that accuracy is the last thing that will determine which medical AI is used. Speed of results is vital. Access to the EMR is vital and being improvisational and daring is desirable. Regurgitating boards answers written by the guidelines writer crowd is a fools errand.

The only thing worse is studying LLMs and writing up a manuscript and sending it to Nature Medicine and waiting 4 months for it to print.

It's like taking a picture of an iPhone on your blackberry and paging your friend that you sent them an email.

So Gemini was best 4 months ago, and CDs were popular in the 90s. Who cares?