3 ms·
> Live caption >I love it for accessibility. I loathe it for privacy. Local, self-hosted, or bust. Same for Live Transcribe. Awesome and scary. Uhhmm... Live
by izacus 7y ago
> Live caption
>I love it for accessibility. I loathe it for privacy. Local, self-hosted, or bust. Same for Live Transcribe. Awesome and scary.
Uhhmm... Live Caption is running on-device speech recognition model. So it's local only.
>Smart Reply
>Some commenters here wrote that this was done locally. I would like some source on this (is training also done locally? Nothing is sent back?). If so, this might be interesting. Otherwise, that's plain scary.
No, the model is built ahead of time from other sources then downloaded and applied locally. Data isn't sent anywhere.
>Everyday Android strays further from the GPL. The way forward should be to help manufacturers play nice (use a few persons full-time for support, promote "mainline" devices to the public as the ones that will always receive updates). I haven't looked into the details, but this sounds like furthering their "trebble" HAL to run (and update) closed-source blobs on an outdated kernel (but who cares, right? because they put less and less stuff in the GPL'ed kernel).
It didn't work for last 15 years, why do you think it'll suddenly start working now?
- MayeulC 7y agoAl right, if those are ran locally it's great. But I take issue at the fact that they make it difficult to tell. And I'd be really surprised if they didn't try to leverage an internet connection when available. A pre-trained model sounds a bit less useful, but that's likely a better use of the energy that would otherwise be spent training individual models. > It didn't work for last 15 years, why do you think it'll suddenly start working now? Well, it seems to be kind of working with chromebooks. The market has changed since then. And Google really didn't do a thing about it for that whole time... Also, legislation is slow to catch up, but I would have hoped for some "right to repair" or similar to chime in at some point, and change the cost-benefit analysis. But it might yet change. Lima, Freedreno, VC4 are slowly becoming usable for the GPU, and there is a slow but steady uptake of devices getting their drivers mainlined. It's a shame that mostly concerns older devices, as people don't necessarily like to tinker with their $1k+ devices.