We really need a few new protocols: Encourage the use of something like an AI.txt in git repos that either gives model builders permission to use the repo cont

We really need a few new protocols: Encourage the use of something like an AI.txt in git repos that either gives model builders permission to use the repo contents as training data, or not. Model builders need to take reasonable care to avoid incorrectly using training data that they don’t have the rights to use. I have been using GitHub CoPilot since the beginning, I love it both with Emacs and VSCode, and I would allow my zillions of GitHub repos to be used as training data. I have published all of my recent books under a Creative Commons License, and I encourage reuse when allowed in derivative works. That said, I don’t think it is possible to get permission from the zillions of web, books, and articles authors to allow their writing to be used for model training - but, LLMs provide so much possible value to society so I think we need lenient copyright and reuse laws. I am skeptical of a few tech companies controlling AI, but my recent experiences using open models have been promising. (I have been running Vicunu 33B for research for my new book Safe For Humans AI). I see a bright future for Open Source (I choose Apache 2 and GPL licenses, but everyone gets to use what they want for their own stuff) and increasingly powerful Open LLMs.