I thought it was hilarious that one chatbot was unable to find any assholes because it was taking the term too literally.

But aren’t some of the stories and responses AI generated?

If you take a dataset and fit a model to it, then use that dataset again to gauge the viability of the model, the model will pass with flying colors - because it is literally fit to the dataset you’re using. You have to use “other” data, even cutting the original dataset in half and using half to train the model and half to test it, is better than nothing.

If much of reddit is bots, then this study would be bots feeding on each other.