Update 2: After excruciating 20 hours or so I finally managed to download all my photos. I cancelled my subscription, shut down my account and wish you guys all the best!
Update: I assume Safari puts tabs in sleep mode after it doesn't detect activity. As long as I keep my screen unlocked and use Firefox, I can download files. Although, they're very slow since there is not an actual export option. In my opinion, Drive needs a lot of work and Proton needs to make it easier to export out of all their apps!
Sorry, I have to vent right now. I am really sad that I will probably lose my pictures of 2024. I have a Proton Drive subscription, but it's been giving me a lot synchronization problems. I won't go into detail on them, because I have already spent so much time with Support to find a solution. There just isn't any and it's unclear they will ever fix it.
So I decided to cancel my subscription on Proton Drive at least. Turns out it is practically impossible to download my media from Proton Drive. This is my sixth attempt downloading media and it keeps getting canceled close to the end. There is no other export option than to manually select media, download them through the browser and pray that it will finish.
I tried with 500 items, as support recommended me. It didn't work. So I worked my way down to 400, now to 300 and it keeps failing. I don't know how to get my pictures away from this terrible Drive app. It's not doable to work in very small batches. I don't have this time on hand and Proton can't offer any solution either. I cannot even downgrade my subscription if it's using up more than the 15GB storage that they offer.
After I find a way to export all my media, I will close down my Proton account. This is not the way of doing business. In the best I managed to easily download all my media from Google Photos and iCloud. I regret leaving iCloud and at this point, just invest the money in self-hosting solution.
Never Proton Again
how is the llm not actually private?
How would you make a private LLM other than one that runs on your local GPU?
You have to send their servers a plaintext prompt for a cloud LLM to work, there is no way around it.
it sounds like they are saying they don’t use your data and encrypt it so only you can read it, i assume the same as they do with protonmail?
But how would that work with a cloud LLM? The prompt be can't encrypted, they have to read it in plaintext. They don't need to read your email contents to make the process of sending an email work.
They can subsequently encrypt it "at rest", e.g. for the duration of the LLM session or for some hours or whatever, but they would have to have access to prompt/output data to make the LLM work. It's not like email.
i don’t think they’re saying that the processing of your llm query is encrypted, just storage of the chat and the lack of your input data being used for further training or sharing (this is quite important to businesses who don’t want their data to leak)
But then you're back to square one. You rely on trust in Proton (which is fine); but it's not really private (in the dictionary sense of the word).
And if you have to rely on Proton, you're in the same situation as with OpenAI/Google/Mistral etc (I am speculating, but I highly doubt an enterprise subscription to say Google Gemini allows Google to leverage the prompt/output data).
Although there is actually one important caveat to the above-mentioned point; the quality of Proton's LLM output is noticeably worse than with say Gemini (I only did some basic work-related prompt tests, but it was significantly worse than Gemini, ChatGPT and Mistral - LLMs that I cross test relatively regularly).
For the vast majority of use cases about anything, we rely on a firm and we don't have an option but to trust it to one extend or another. Very few things out there can operate with E2E which isolates us from this trust. This is something often missed by peope in the tech circles, as well as the corrollary that it's the firm that needs to be improved and not the technology, because often there is no practical technological solution. In this regard Proton being a non-profit is on a different level compared to the publicly-traded profit-maximizing multinationals like Google etc. I'm not saying one has to have complete trust in Proton. Instead that one is in a very different boat using a Proton product compared to a Google product.
yeah to me it looks like just a consumer targeted version of enterprise versions of gemini/chatgpt etc
openai looks to have a simple option to turn off using your data to train so the only truly unique thing appears to be encrypted storage of the chats
yeah they got a long way to go, im still using mistral and claude mainly