You know that's not the LLM's 'source' right? It's still output. Do you mean the training data? Is that what you mean by CoPilot should be open source? If CoPilot has learned from something GPL then everything else it outputs, or perhaps specifically its training data - should be GPL?