8
Ran into an old professor at a diner and he dropped some AI wisdom on me
I was eating at this diner off Route 9 in Albany last Tuesday and saw my old computer science professor from like 5 years ago. He told me the biggest shift he's seeing in AI is how small models are beating big ones for specific tasks now. Said he trained a model on 200 patient records that outperformed a giant hospital system's AI for predicting readmission rates. Has anyone else noticed this trend with smaller specialized models?
3 comments
Log in to join the discussion
Log In3 Comments
stellaperry25d ago
Wait @william_harris trained on only 500 tickets? That's wild it actually worked better!
1
william_harris26d ago
Yeah, I used to be all about bigger being better with AI models. But your professor's example about the 200 patient records changing my mind. I tried something similar at my own job, trained a small model on maybe 500 customer support tickets and it caught patterns the big generic model missed completely. Smaller models are way cheaper to run too, which is a big deal for most people.
0
spencer_coleman26d ago
The thing nobody talks about is how smaller models actually force you to clean up your data better since they can't brute force through the noise. Did that change how you organize your support tickets now?
2