While visiting Boston, I asked a doctor and a drug discovery researcher about how they use AI. Their answers made my day: AI is used outside the tech bubble more than I expected (and in the right way).

I asked the doctor if she uses LLMs to aid in diagnosis. I expected a "no", given how backward the medical field can be when it comes to technology (this same doctor has to send FAXES for every patient visit).

Surprisingly, they have their own LLM fine-tuned on peer-reviewed medical papers (openevidence .com), and it apparently works pretty well. She uses it even when she's pretty sure of her diagnosis, just to check she didn't miss anything.

The way she talked about it raised all kinds of green flags based on my own experience with using LLMs. For example, she doesn't trust it blindly and relies on her own training as a fallback.

I asked the drug discovery researcher if the AlphaFold hype is justified. AlphaFold is impressive on paper (e.g., in academic contests), but I've heard people question why there are no major real-world results attributed to it, even after earning a Nobel prize.

He uses it every day and says it's legit. The reason why there are no headlines like "X illness solved thanks to AlphaFold" is that it only helps with one of the steps in drug discovery, but it helps a TON for that. It's just not something designed to produce end-to-end results.

This got me thinking about how other fields are using AI. If you have first-hand experience, please share!