[jauntywundrkind]

How can this be done without a standard? By having a user download a couple GB model for every site they visit, and using webgpu (until/if web-nn ships)?

This doesn't make any sense. And it's far less ideal than using already on device models, that are better tailored to the device's hardware, and that may be already resident.

Good things should be made broadly available. We didn't need jquery before it came along, we could just do the things. But jQuery set a standard. And in turn much of what it did got sent into improvements to query selector. Your model of standardize only the minimum sounds like suicide for the web and I think it's dreadful, and I'm glad such dour sad views have no standing.