@davidrevoy I'm not sure so far. ML has valid applications, and so far what Mozilla has brought to the table seems very reasonable to me. Most important thing for me is that everything is local and privacy friendly. And the features they have are very reasonable: An alt-text generation model makes a lot of sense for people to whom alt-texts matter. And local translation of websites also sounds like a really nice feature. As far as I see such features will make the web more accessible with better privacy.

For me the most important things are:

1. ML should be local and not send arbitrary data to shady third party services
2. ML should be build to only affect performance when you actually need it
3. ML should provide specific tools, not general purpose blackboxes

Also
4. ML should be trained in a responisble way, which means:
1. Responsible source of training data
2. Resource use of training should be justified by the benefit of the functionality

I remember the time when ML was mosty an exciting University thing, where results were shared, and not some big corporation BS. I am not against going back in that direction.

But you know, digiKam has been using public models for facial recognition and matching for years now. And I don't think anything bad came from it.

I am currently in the position of saying: Wait and see what mozilla does. And if it turns out the wrong way, there are forks. That's the beauty of open source.