0

I wonder if it would be legal to train AI code generators on various code bases separated by license and then get assistance on your project by whatever model complies with your project's license.

This question differs from Can I use part of a MIT-licensed library as training data for an AI that will be used commercially? in that the latter can be answered (and has been answered) without answering my question: Obviously you can use the generated code commercially regardless of how licenses transfer onto it because MIT allows for commercial use. What exactly the situation is in terms of licenses is irrelevant to that question but is the core of mine and it remains unanswered.

1
  • 2
    Sorry, it's a duplicate. The dupe makes it clear in the accepted answer that either the model's outputs aren't a function of the inputs, in which case you can license them how you like, or they are, in which case the output needs to be distributed in a manner consistent with the requirements of the MIT licence on the inputs - which would definitely include "distributing under the MIT licence". It's true that the linked dupe doesn't explicitly answer your question, but all the information you need is there - and we do require a little thought on the part of question authors.
    – MadHatter
    Jan 20 at 11:19

0