3 ms·

I assume this is all done server-side, with Google feeding your interests and investigations into your personal profile. I really wish this could be done on-dev

I assume this is all done server-side, with Google feeding your interests and investigations into your personal profile. I really wish this could be done on-device. I don't want Google slurping up even more data about web users. I use FF mainly on desktop and mobile, and there's probably a great opportunity for Mozilla to build an offline, privacy-first summarization model.

hgsgm 3y ago
Yes, what Mozilla needs to do remove more support from browser development and debugging, to fracture their efforts even further, to continue the line of successes that began with Persona and continued with Firefox OS.
ttul 3y ago
I’m wildly speculating, but it seems that projects like llama.cpp are bringing SOTA models closer to the desktop. It’s only a matter of time before browsers embed a small LLM for various purposes, providing access to this local model via a JavaScript API that allows clients like Gmail to perform tasks on data locally. Apple would be strongly incentivized to do this, given their value proposition of user privacy.
- fidotron 3y ago
  I think the fact Google curiously ignored all the security problems raised by the WebGPU API suggests they are closer to trying to offload the GPU inference part of this to end users than people think. Build as much of the model as you can in the cloud, run inference locally and push results back is probably the cost optimal way to run this stuff at scale.
DoingIsLearning 3y ago
Kagi has been doing this already for quite a while, well before google (see Kagi Universal Summarizer) with no info linked to your profile [0] [0] https://help.kagi.com/kagi/getting-started/faqs.html#i-have-privacy-concerns-over-linking-my-search-queries-with-my-credit-card-why-should-i-trust-you https://help.kagi.com/kagi/getting-started/faqs.html#i-have-... For the record I am a happy user, not affiliated.
lxgr 3y ago
> I assume this is all done server-side I assume so too, and given that, it's incredibly frustrating (but not at all surprising) that they require users to use Chrome to be able to use the feature.
bradly 3y ago
At Apple, the challenge our team faced with on-device models was managing their size and update-ability when every team has multiple models.
judge2020 3y ago
This doesn't change how or when Google sees the web pages a user visits. If they were using Chrome with all sync features turned on, their browsing history was already being sent up.

hgsgm 3y ago

Yes, what Mozilla needs to do remove more support from browser development and debugging, to fracture their efforts even further, to continue the line of successes that began with Persona and continued with Firefox OS.

ttul 3y ago

I’m wildly speculating, but it seems that projects like llama.cpp are bringing SOTA models closer to the desktop. It’s only a matter of time before browsers embed a small LLM for various purposes, providing access to this local model via a JavaScript API that allows clients like Gmail to perform tasks on data locally. Apple would be strongly incentivized to do this, given their value proposition of user privacy.

fidotron 3y ago
I think the fact Google curiously ignored all the security problems raised by the WebGPU API suggests they are closer to trying to offload the GPU inference part of this to end users than people think. Build as much of the model as you can in the cloud, run inference locally and push results back is probably the cost optimal way to run this stuff at scale.

DoingIsLearning 3y ago

Kagi has been doing this already for quite a while, well before google (see Kagi Universal Summarizer) with no info linked to your profile [0] [0] https://help.kagi.com/kagi/getting-started/faqs.html#i-have-privacy-concerns-over-linking-my-search-queries-with-my-credit-card-why-should-i-trust-you https://help.kagi.com/kagi/getting-started/faqs.html#i-have-... For the record I am a happy user, not affiliated.

lxgr 3y ago

> I assume this is all done server-side I assume so too, and given that, it's incredibly frustrating (but not at all surprising) that they require users to use Chrome to be able to use the feature.

bradly 3y ago

At Apple, the challenge our team faced with on-device models was managing their size and update-ability when every team has multiple models.

judge2020 3y ago

This doesn't change how or when Google sees the web pages a user visits. If they were using Chrome with all sync features turned on, their browsing history was already being sent up.