Email or username:

Password:

Forgot your password?
Top-level
Ryan Fisher

@brhfl
@sertaptap @davidgerard

The two blog posts by the actual ML head at Proton seem a bit more circumspect than the hype.

I thought the post on using LLMs in a privacy-focused design was at least thoughtful and had a lot of interesting information in it.
proton.me/blog/how-to-build-pr

The announcement post on Scribe does point out that the whole thing can be run local etc. It also seems likely they are trying to work out the future of the product with some of the more advanced privacy ideas like homomorphic encryption (an idea which still hurts my brain).
proton.me/blog/proton-scribe-w

2 comments
bri

@wryanfisher my takeaway was more 'damn that's a lot of resources spent on something less than a third of your customers give a shit about' but w/e. was just here to relay the survey info, i already know i'm not giving them any more of my money.

Ryan Fisher

@brhfl

I think that's a totally fair reaction.

I feel like the opt-in affirmative action to enable it on a per use basis mitigates a lot of the privacy issues for me, though. For the writing assistant, at least. The LLM is not training itself on your mail corpus I don't think.

Go Up