I am currently using Perplexity Pro for research. I wish to upload certain personal and confidential pdf files for analysis; however, I read a few articles which highlight the privacy concerns of perplexity. I do not have PC with a good hardware config for processing the pdfs by installing local LLMs. Do all the AI platforms online have similar privacy concerns as that of perplexity?
Anything that you submit to any GPT is neither private, nor confidential anymore. Irrespective of what it says, unless of course you are doing it entirely locally.
1 Like
Nearly every AI platform needs to use every bit of data to retrain, so nothing is private unless you run it locally.
It is also the reason that most established enterprises don’t use off-premise solutions for AI or any critical business functions.
What kind of specs your PC/desktop has? While some complex models require beefy configurations, there are some low quantized models which can run on low-end devices too.
Another option if you need absolute privacy is to rent a GPU in cloud. You can use some options like runpod.io where you can get a 24GB VRAM RTXA5000 for $0.26/hr.
Absolutely right, I agree 
Unless you are hosting it, it’s not private.
You could technically use Open AI Platform and under data use you “check” the box that says not to retain data and not to use for training. I think they delete it in 30 days. You can also delete your acocunt.
Or than that, if you want fully true privacy, you need to run AI Models setup in your own GPU.
Even renting a GPU in cloud may leave some unknown traces.
FWIW recently this happened: https://openai.com/index/response-to-nyt-data-demands/
Ye what ever you upload can be used for training and can be used in a indirect way, my company has hosted an on Prem version of the AI tool, just to implement HIPPA and all for client side.
So they’re claiming that they’re serious about privacy by not storing the data forever. It appears that there’s no end to end encryption of our data. But other than that there’s no access given or distribution of our stored data to any third parties. There’s also an option to disable the training of AI models with our data. If we disable that, wouldn’t there be a decent level of privacy?
https://openai.com/en-GB/policies/row-privacy-policy/
That is more of a sensationalist article posted by them to fight against the copyright case ongoing with NYT.
Instead, if you read their privacy policy, they can keep and share the data with any of their partners indefinitely and may or may not de-identify the data. Basically they can retain and reuse any interaction you have with them, directly or indirectly.
Privacy is not only about model training but how they retain and use the data. In that sense, they can retain, share or even sell your data, while not committing to anonymising it.
1 Like
Sorry to repeat what others have already said but nothing is private online. Your only bet is asking someone(friends) with better GPU or RAM to do this for you.