> I think it's interesting to see in which ways the models fail and that they actually produce some good stuff once in a while.

This is a good summary of GPTs.