Post #4190826

We know for a fact that LLMs have been trained on all available open source codebases. Do we know that there is a relevant amount of closed source code in the training data?