a great piece by Joseph Gordon-Levitt on how we might start thinking about IP and residuals in the world of AI-generated content.
I think the hard bit is possibly not the tracking of training data through the generative process (hand-waving some deep technical problems) as much as it is the funnel of attribution and payment. tracking this, maintaining it, is a whole other scale of organisational complexity.
creating a programming language brings into clarity all those seemingly odd and arbitrary constraints other languages have.
like how to define functions in your assembly language but not have the execution engine (or VM) pile through them as if they were part of the main flow.
thus the "globals are special and here's the well known entry point" rule is born.
the power of #rust in action.. an upgrade from nom 4 to nom 7 (yeah, I know) brought with it some breaking changes. like fundamental changes to the API and integration surface.
I thought I was looking at a full rewrite for my pet language (compiler, assembler, VM, repl) but it took all of a couple of hours, with the support of the robust testing, logging, and borrow checker frameworks.
in any other language this would have been a burn-it-down moment I think.
police using facial recognition at scale here. the moment a chant crosses a line into crudeness or mild aggression the cameras come out and the tracking starts.