I am not as optimistic about the real value of this as you are, but my practice is, ironically, IP and general advanced software counseling specialized in ML and decentralized/distributed network tech, so ymmv. Nevertheless, I think the increasing prevalence bodes poorly for the industry and consumers for two very specific reasons. For the industry, a major part of making good lawyers is by having green lawyers be shit lawyers under close supervision and lots of repetition. The rise of legal AI has already totally short circuited that fundamental process quite a bit. The other concern i have is that consumer reliance on these technologies actually pressures legal systems to converge towards greater convolution and arbitrary heterogeneousnes. This is bad because it (a) makes human inspection less and less feasible, (b) introduces a great many failure points that we’ll never be able to discern until it’s too late because of (a), and © makes the surviving equivalent of practitioners (and even automated systems) more vulnerable to the myriad issues with hyperspecialization. It is noteworthy that these are very similar issues facing the software industry.

So the admission of intentionally circumventing the NYT paywall is where this becomes a real actual IP infringement issue for them. From a legal perspective, that is literally as crazy a thing to come out in discovery as Anthropic’s torrenting an enormous chunk of their training corpus. These are IP infringements, totally irrespective their use in training GPTs. I cannot imagine having to represent these idiots.