I have thought about writing a python web framework were instead of writing a function that handles a request, you write a docstring and the an AI JIT generates your handling code. Could we not just prompt-engineer a solution for the missing bits in the driver for the HDMI2 stuff and have it be lazily generated via a model parameter and an API key? And then, in about 10 years, we could just do it locally once the models become runnable on commodity hardware. What a future too look forward to.
Changes are the leaked spec might already be in the popular LLMs' training data, so you could probably skip to step 2 without having to do the (potentially problematic) training yourself.
IP vs AI, round two, Fight!!!