I think there's a real possibility this is a "no such thing as bad publicity" stunt.

Those two are not linked. I could buy that maybe human-readable code will be the minority.

But what does ephemeral code even means? That we will throw everything out of the window at every release cycle and recreate from scratch with llms based on specs? That's not happening

I can answer that one: none.

The only thing I can think of is massively increased context windows (around 4k for gpt3), but a million context token with degraded performance when full is not what I'd qualify as resolved.

It might not have been. But it's not hard to see that whatever productivity coefficient multiplier llms brings, it's being dwarfed by how much easier it is to publish projects that only look good on the surface.

While it's a great tool in the hands of capable and well intentioned people, there's not a world out there where the average quality of software goes up.

I think you very much want to code differently if you were to target gddr instead of ddr. Unless you were only working at a high-level of course, but it would significantly influence eg the rendering engine