235
Google will use Gemini to detect scams during calls | TechCrunch
(techcrunch.com)
This is a most excellent place for technology news and articles.
From the article it's entirely on device.
For now.
It's probably too expensive for them to record and upload every call, but im sure the transcripts of calls are likely to leak soon after this.
Yep , we will let your hardware do the work , we will just collect the final profile data that's valuable to us.
This is genuinely really cool.
The best/worst part of the AI boom for me has been waiting for the advances to trickle down in terms of open source models and on-device models, rather than having to send everything up to the cloud.
Obviously this isn't an open source model, but the on-device processing is great.
I’m worried about the security of this. However looking past that and simply observing it as an implementation of AI, this is an idea that I think is actually a beneficial use. Protecting the elderly against fraud/scams is a major issue which gets increasingly complex as scammers improve their methods. Using AI to detect scams in calls could be helpful in protecting the elderly.
But before rolling it out, I would want to see proof of its efficiency through careful studies. Hell, incentivize Google to share the model with the government and other businesses so it can be improved upon. Fund it as a grant/program so smaller teams/companies can contribute and innovate.
They do have versions of the Gemini models that are open sourced, called Gemma. So far they don't work quite as well as the Llama ones from Facebook but I'm sure that will change.