Sunshine (she/her)Proton Launches Private LLM Called Lumo
katy ✨god andy yen is just the worst; he continues to ruin proton any way he can.
anonfopyapperWhat is wrong with adding an llm? That is not even integrates in other proton products?
EU needs in-house LLMs
llms are rubbish and shouldn't exist
iktgreat! then why are you here?
please let the adults in the room have a discussion
Adutls be 🤡
Proton is based in Switzerland. Not EU.
They said it integrates with Proton Drive. Optional, but still. That is data that is otherwise strictly end-to-end encrypted, and now they’re adding a “convenient” method to send it through their AI unencrypted, and they are not upfront about that.
Proton has a problem with focus. They keep adding new things of limited value, presumably in a quest for growth. That usually doesn’t end well.
We are still waiting for the contacts integration and proper drive functions
Yep. They have their summer roadmaps up, which include a Proton Drive SDK and a Linux app. Hopefully the SDK will open up more possibilities for the open source community.
I understand your point of view, but I think that in today’s society, a company can’t afford to fall too far behind when it comes to current uses. I’m perfectly aware of the risks that AI brings, but if it can enable some people to find a -private- way out of American AI, I think it’s a good thing.
it can enable some people to find a -private- way out of American AI, I think it’s a good thing.
Lol what?
There is nothing private about taking ZK encrypted data and feeding through another man's server unencrypted.
Educate yourself please.
Hard agree, I’m sure he’s sending all your data to Putin
Andy is a grade A bootlicker
privacy first
model is not open source
trust me bro we don’t share your data
It’s the Mistral Large model. Mistral does open source many of their models but not the large.
It’s always been a honeypot
Proton is dead, who cares.
No dark mode 👎
zonnewinBut Dark Reader works fine.
Did Proton train their own LLM or is it a Mistral model?
Sure enough
Lumo being a cat is a cute nod to the Mistral team, who’s AI chat bot is Le Chat (the cat).
The second time I had to try a little harder to get it to admit that it was Mistral.
Kind of light on details. “Lumo is based upon open-source language models”. Okay. Which ones? [Edit: they offer more details at proton.me/support/lumo-privacy : “The models we’re using currently are Nemo, OpenHands 32B, OLMO 2 32B, and Mistral Small 3”]
Not sure how I feel about this. I figured Proton would find some clever way to run models on encrypted data, or at least do something akin to Apple’s “private cloud compute” but…nope, just another cloud platform like any other. Zero-logging is all fine and good, but don’t pretend like you can’t access my chats when the only thing stopping you is your logging policy.
Web search — If you ask it to, Lumo can search the web for new or recent information to complement its existing knowledge.
Again, no details. So you’re not sharing my data, but you are potentially leaking it to unnamed search engines? Cool, cool.
Oh, and it has built in Proton Docs integration, in case you wanted to accidentally send your documents, unencrypted, through Proton’s servers. And also maybe leak their contents to a third-party search engine, who knows?
Please, Proton. E2EE or GTFO. The world doesn’t need another chatbot. If you can’t do it right, just don’t do it.
Lumo itself claims to be based on Mistral’s Large model, which is not open weight.
lemmy.world/comment/18393337
Agreed, the lack of E2EE is a huge miss for a company that built its reputation on privacy - running open models like OLMO and Mistral is nice, but without true end-to-end encryption, your prompts are still visible to them, which defeats the whole “private AI” marketing thingg.
And to where exactly are the servers moving? I don’t trust Proton with this.
why do they still have a Twitter link?
MAGA baby!
Andy the bootlicker lost touch with the people he serves
He should go kiss the ring of the pedo king again
Pathetic
Ever since Proton went MAGA their presence has just gotten weirder. This is another giant step in a direction nobody wanted.
Glad I dropped them.
Alexanderdisgusting
BaŝtoI don’t get it? This is more about a chatbot service than about LMM, right? It doesn’t sound like they release the model, which would mean it’s less privacy-friendly than running R1, Qwen3, Llama, Mistral or Gemma3 on your own machines. Google and Meta honestly offer better privacy with Gemma and Llama.