So - I don’t think Firefox would be generating captions for PDFs on PDF creation.

I'm not sure. The blog post is not entirely clear on that.

Between the document creator and the creation software or screen reader, only the document creator would really know the context for the image.

Agreed. Context is usually very important for images. But with an auto-generated caption embedded in the document itself, you already lose some context. Because if the automatic caption is incorrectly stored as "The ___ video game installer" you cannot decide anymore if this was written by the author with the context in mind or just generated. Which I would argue is worse than no caption, as it lowers your trust in all captions.

But I wouldn’t be opposed to ML on those that can auto-suggest things or even critique how content authors write their descriptions.

Absolutely, I think that will be by far the best solution. It could massively encourage users to write their own captions if in most cases you only need to accept the suggestion. But so far, that seems unlikely to be the way forward. Why do that when you can just throw even more "AI" at the problem?