I confess this is not specifically Fedora related but I don’t know of a more knowledgeable and trustworthy bunch of people to ask other than here.
I now need some kind of paid AI plan, just for general research and image generations. I am concerned that all data is retained by ChatGPT (even if they try to fudge the issue with ‘not used to train models’ privacy settings). I want NO data retention, on principle.
Just wondered if anyone knows if that’s even possible?
Thanks for any thoughts
Unlikely, I’d say. If any of these companies can get hold of more data they will.
They already ignore robots.txt files, hammer the daylights out of servers rather than “playing nice”, don’t respect requests to not scrape and so on.
I’d say the moral compass of “tell the user we don’t keep their stuff and really don’t keep their stuff” is spinning like a top right now.
If you have powerful enough hardware you can run Stable Diffuse on your own machine. My son run it on ab Nvidia 4090.
Not a solution if you want the benifit of a fully trained LLM.
Maybe Lumo (the LLM assistant by Proton) fits the bill?
Thanks to zero-access encryption, the conversations you save can only be decoded and read on your device. Proton can never see them, and neither can anyone else.
… or so they claim, at least 
You can be sure data is kept and you are watched and recorded. It is not ChatGPT, it is all the current “platforms” on the Internet, the Operating Systems (see Windows) and the hardware like the phones. Your only hope is to blend in the background of billions other people being watched. Which of course becomes difficult as soon as you are somehow different for a reason or another.
Thanks, I heard something similar but unfortunately I do need the full trained models for web research etc
I actually found that today, I like the privacy angle but it doesn’t have web access and can#t do images etc, sadly
I guess it’s back to subterfuge then. Cash prepaid credit card, VPN usage, no personal data entered (although keyboard fingerprint is pretty hard to get around, but I can do what I do sometimes already, type in text doc, ask one AI to rephrase, then paste into another :D)
I see generative AI as mostly a load of bollocks, so my opinions are probably not worth much to you. Still, I’m curious to know why you think it’s beneficial for research over a good old search engine. For image generation I suppose it can be mildly useful if you just need a quick and dirty way to generate an image of a thing.
Couldn’t agree more, except I’d say it’s a bit more ‘significant’ than ‘bollocks’, positively Satanic would be more my opinion, but then I felt (and still feel) the same way about Facebook and all (anti-)social media.
But with competitors using it, and doing what would take me hours in minutes, It’s becoming one of those ‘can’t refuse to use and stay in the game’ things.
I tend to use it in very specific ways though, and it’s usefulness can’t be denied, certainly not comparable to using search engines (which will be a thing of the past soon anyway).
Example: for years we have had horrific electrical problems. Can’t afford electricians to spend hours fault finding in such a complex and old setup. I used GPT for 30 minutes and think I have solved it in principle. Then I could ask questions no electrical forum will answer (i tried), due to the nanny state rules and laws, latest 18th amendment regulations, building regs, etc etc. So I asked GPT to read all the local and national laws/regs etc, then to advise me on which parts of the fix would be legal for me to do without a sparky certificate. It did it superbly, even recommended actual products to fit dimensions etc, then told me which bits require a sparky to stay legal. In 30 minutes I solved a problem that has plagued us for years, and in that time I spent probably 50-100 hours on search engines, forums, to no avail.
Granted, I detest ‘creative’ use of GPT, i.e. lazy f*cks charging $50 for an article for a website which they got written in seconds with a software program and API credits to AI models. Although for images (let’s say logo design ideas), it’s pretty remarkable at what it can achieve, but again I only use it for ideas/direction, never to just copy what it produces.
To each his own, it’ll kill us all soon enough anyway 