I think people who seem overly eager to call chatGPT and stable diffusion and similar models "bullshit generators", and dismiss them because of that is missing something large.
In Machine Learning itself, we consider two classes of models/algorithms, discriminative and generative. Up until six months ago, the best work we had in ML was all discriminative, and humans were irreplaceable when it came to content generation. All of a sudden we have good generative models and a way to cheaply generate vast amount of content and ideas, some of which are good and some of which are terrible. I predict this shift will also shift the associated humans' roles from "generator" to "discriminator", who will be cherry-picking and editing AI generated articles or images (much faster) rather than creating them from scratch (much slower). Yes, it will take a slighlty different set of skills in the human worker, but if you can't see how that productivity boost can help humanity, I ont know what to say.
In Machine Learning itself, we consider two classes of models/algorithms, discriminative and generative. Up until six months ago, the best work we had in ML was all discriminative, and humans were irreplaceable when it came to content generation. All of a sudden we have good generative models and a way to cheaply generate vast amount of content and ideas, some of which are good and some of which are terrible. I predict this shift will also shift the associated humans' roles from "generator" to "discriminator", who will be cherry-picking and editing AI generated articles or images (much faster) rather than creating them from scratch (much slower). Yes, it will take a slighlty different set of skills in the human worker, but if you can't see how that productivity boost can help humanity, I ont know what to say.