this post was submitted on 22 May 2024
80 points (98.8% liked)
Firefox
17906 readers
32 users here now
A place to discuss the news and latest developments on the open-source browser Firefox
founded 4 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
If AI object/scene recognition is done locally, wouldn't it increase the memory footprint of the browser process. Also how many objects can it identify if its run on a modest 4-8 GB RAM system? One more question is would they ever introduce anonymised telemetry for these generations?
The state of the art for small models is improving quite dramatically quite quickly. Microsoft just released the phi-3 model family under the MIT license, I haven't played with them myself yet but the comments are very positive.
Alternately, just turn that feature off.
If it works anything like Firefox Translations does, the model is only downloaded on-demand, so it wouldn't affect your browser usage if you don't use the feature.
to be faiiiiiiiir, the way they’re going about it is very reasonable. i’d rather have no AI but, if i had to have it, i’d rather have that than anything else.
The use case they mention (generating alt text for images in PDFs) is something that couldn't work otherwise and, even if it isn't perfect, can be a big help to people with visual impairments, while at the same time doesn't get in the way of the users that don't need it.
If they keep focusing on these kinds of features instead of going fully Clippy like Google and Microsoft are doing, I think it's fine.
honestly, you’re right. I still worry that it could encourage an attitude of abled people not caring about alt text, because "oh well AI’s gonna do it anyway, who cares!", but, really, abled people already don’t care about alt text, so…
In the specific case of PDF most users wouldn't even know where to add an alt text. Depending on how you generate the PDF it might even be impossible. So I think Mozilla has the same concern as you, and that's why they aren't adding this to images in HTML (yet).