The Importance of Fine Tuning Language Models

There's also a lot more in the pure language model world beyond scaling up the existing stuff. I still think we have a long way to go and fine tuning the models in a smarter way. We definitely do get what's called mode collapse or entropy collapse where in some cases the model will output a very narrow set of answers or a single answer. If you ask Chatubt, tell me a joke, it'll probably always tell you the same joke.