Agree with this framing, but we may find out that the answer is somewhere in the middle. Personally I think it is extremely unlikely that this is straight plagiarism in the sense of the model simply regurgitating training data from prior work by Tristan, but it is very plausible that his work (and others’) was foundational to the breakthrough. What is unfortunate is that the stakes are so high (and no I don’t mean $1M) and the timeline is so compressed.

I expect a lot more of this kind of drama in the near future.