

5·
3 days agoIt’s a recipe for model collapse at the least. Everything will trend toward the mean, and the models will get worse instead of better.


It’s a recipe for model collapse at the least. Everything will trend toward the mean, and the models will get worse instead of better.


They’re also trained on each other’s outputs. I forget exactly which two models it was, but there was an example where, e.g., if you asked Claude about itself it would confidently declare it was ChatGPT.
1/1/00 baby!