Hey all, until now I’m being super reluctant about using any AI tool. On top of all the privacy issues it obviously comes with it (I can’t believe there are people talking to ChatGpt and literally teaching them how to impersonated!), I have a feeling that it only does jobs that you are capable of doing with other tools, but since it looks like a person and not a computer, people start to fantasize that it will do everything they don’t know or don’t want to do it and become numb and stupid. But, maybe there are some uses for it and at my office we will soon have access to Claude, so I want to explore it and check for myself. My question is what precautions should I take and what limits should I put on my usage? I thought about creating a new email account just for signing up, but obviously that’s nothing nowadays.
The only place I find it useful is in writing documentation for the code. I can quickly review it without there being any gotchas that only show up under certain circumstances like with code and even if I miss something they can reach out to me.
There are plenty of innocuous uses. I’ll use it to find the name of something I can’t remember or as a thesaurus.
I run ollama locally and sometimes lumo
EASY GUIDE ON USING AI:
step 1: dont.
so edgy
Run your own. Doesn’t require internet connection. Try downloading a model like gemma 4 and run it on android with pocket pal
You can redact sensitive data before sending it, there are tools for that, i know one named privatiser.
Best to not create an account if you don’t have to, ppq.ai and NanoGPT are proxy services that claim to protect your privacy. Or use the free tier of big services with a vpn and the usual privacy tools.
Other than that, explore, enjoy and don’t tell it anything you wouldn’t tell a policeman. Ask it about a bird you have seen recently, or a phrase in a foreign language that has always confused you. They’re good in things that are hard to search the normal way.
Unless you’re running a model on an air gapped machine that will never connect to the internet again, there is no privacy preserving way to use so-called AI today. All the providers will tell you it’s no problem. But then you read the news about which model fucked up what today. And it’s a lot. Anybody using so-called AI today is voluntarily participating in a massive, not well organized beta test. At their own jeopardy.
So don’t give it your medical history and don’t talk to it about your innermost thoughts. Try to keep it out of your internet browser and history if you can.
host your own:
*needs beefy hardware
If you want real good intelligence (self host ain’t cutting it for most things), then use an ai anonymizer like duck.ai or lumo(haven’t tried this one so idk). You can also use chatgpt.com with TOR. I would just keep in mind no matter what your prompt are being trained on, so don’t provide identifiable info.
What’s your threat model and what do you want the AI to do?
Does it need an internet connection? Do you have to use a service rather than the dataset/model? Do you have a machine powerful enough to run it locally? Is it for work or personal use? Why do you have to create an account to use your work’s Claude subscription?
for starters, don’t use Claude Opus to do any real work. Opus is good for planning, but once it has a plan, switch to a cheaper model like Sonnet to do the actual work. using Opus for everything is how people burn through all their tokens in a day.
Try Lumo from proton if you want to stay private. I was (am) reluctant too, but i found out it can be useful if I am writing something more official, NOT to let it do it for me, but the sentences I am not sure about, it can tell me if they are fine and explain why (and you can verify with easy search because you know now what you are looking for). It’s especially helpful with my non native languages. Otherwise i did not yet find any other useful use case and I refuse to use it for dev.
