Pretty interesting and helpful video about an important topic which isn’t covered nearly enough in my opinion. Also related thread about adding AI Tools to PG here.
2 Likes
Sorry for derailing the conversation. This is the third time I have seen content by this creator getting pushed on PG. Is it really that good?
This is the first topic I’ve made for one of his videos but I’ve used them as references before. I find they are well researched (see sources in the description), and thorough while remaining accessible to the masses (easy to grasp.) Obviously this post was a swing and a miss 
on my part but I thought it could be helpful for people here who have been asking about using LLMs privately.
The biggest valid complaint which I can agree with is that he has a tendency to tangentially mention GrapheneOS when it’s not super relevant to the video. (This is not a crititism of GrapheneOS, please don’t take it as such.)
I’ll add that nothing on this forum is endorsed by PG but personally at least I would generally recommend his videos.
I agree that the last part about GrapheneOS and using separate user profiles is a bit odd for a video about AI.
Saying, “What if you need even more privacy? What if you can’t make any compromises whatsoever?” after mentioning running local AI-models, arguably the most private way to run text and image-generative software, is weird. I don’t quite understand the need to go so far as to buy a Pixel, flash GrapheneOS, create a separate user profile, enable always-on VPN, and pay for the account with giftcards in order to dissociate your identity from your prompts, if your prompts are that sensitive. They can still potentially be stored and used by whatever third-party you’re using. Just keep them on your machine and offline.
off-topic
Granted, I haven’t used LLMs extensively, so I don’t know how good they truly are at processing complex prompts and not derailing off the questions you ask them. But I struggle to see how even the current best LLMs can be good therapists or business advisors. You should not ask computers such things and seek qualified professionals.
I didn’t watch the video, but judging by the chapters, this seemed to give bad advice. Using an online chatbot with an account - even with VPN/TOR and an allias email- will NOT be private. Why ? Because at some points, you will end up leaking information.
Also, super disappointed he didn’t talk about duck.ai (by DuckDuckGo) which is free, proxies your chats, doesn’t need an account - and give you access to very powerful models.
He encourages creating disposable accounts using aliases. However I can agree that over TOR this should really be done with disposable emails that aren’t linked to you, not though a service like SimpleLogin.
I am also not sure why no one mentions duck.ai, I haven’t seen to mentioned in any other video or discussion on this topic I’ve seen. I’ve been using it for a while though and generally find it to be very good.
2 Likes
With SL it can’t be associated with you AFAIK.
But still, your different accounts can easily be linked back together. Unless you are very strict with yourself and create one every day, this is just a sloppery rope.
And there is good chance you will get banned. I myself used Anthropic with an allias and VPN, and they banned me for no apparent reason.
If you have a single SL account from which you create all your aliases which you use across various TOR sessions, then SL could associate all those aliases with your SL account, and thereby your anonymity across sessions would be compromised. Obviously, this isn’t a concern if you only want to prevent the AI companies from knowing who you are.
I use most commercial LLMs using Hoody AI, and sometimes Openrouter but it’s not Tor friendly.
I’m considering self-hosting, saving-up for 2 4090 right now
I mean, SL sure could, but that will kill their business. But even IF they did, that wouldn’t tell them much except ‘This guy really want private ai’. OFC, they could sell the info to the companies but that’s such a strech