GPT-3 is able to accurately predict a lot about COVID-19

Originally published at: GPT-3 is able to accurately predict a lot about COVID-19 | Boing Boing

It may be tempting to assume that there’s something more useful going on here than there is, and the source for this article has clearly has a vested interest in suggesting that machine learning is generally useful, but their tactic of calling the GPT-3 output “predictions” really doesn’t sit right with me. Whilst a model trained specifically on medical data might be able to add useful output, GPT-3 is purely language-focused, and any resemblance to actual reality in its output here should be viewed as largely coincidental, though it was primed and prompted specifically to illicit more relevant output.


It predicted that vaccine would be ready by the fall of 2020, which is pretty correct and not something many experts accurately foresaw.

I wouldn’t say it did better than the experts there. Fauci was saying “end of the year” for quite a while. And right from the start, 12-18 months was the given time frame. “The fall of 2020” can mean September, and it wasn’t approved until December 11 which is winter to any rational person.

If you make lots of random predictions, some of them are going to be right purely by accident. Then, if you only focus on the ones you got right, and, boom, you’re a genius!


This topic was automatically closed after 5 days. New replies are no longer allowed.