Is There a Way to Self Improve?

I think it's a very interesting world here where, you know, language models and NLP are starting to look a bit more like the AlphaGo world. You can use that to build a new training data session and like retrain the model to improve its accuracy. It's almost like you're not multiplying out the probability of failure. But I think these things feel like qualitatively different capabilities than we've had.