The protein folding model was sort of a precursor to the LLMs we have now and advancements in LLMs may yet carry over to such models in the future too. It was also very much a big corporate thing, it was made by Google.

I have little against the local translation feature, especially given the fact it isn't bundled but a seperate download.

That being said, there are higher priorities, like better history and bookmarks, because the current sidebars are 20+ years old and it shows. Chrome has had it figured out for 10 years straight now.

The right way for AI features would be to partner with outside orgs and projects. As you said, they're a waste of development time. And given how much low hanging fruit there is, there's no justification for following AI trends instead of doing basic UX hygiene.