You Really Need to Try Private AI

I have seen reports that current AI is a good personal coach, teacher, guide, mentor, therapist, etc.

But I never paid much attention to this. I just didn’t think I’d be open enough to benefit. Asking intimate and private questions, reflecting on my own weaknesses, revealing my deepest thoughts and fears, uploading journals written in private. None of that seemed fit for any of the current AI offerings.

So instead I continued using AI for data retrieval and coding.

But about a week ago that all changed.

I finally tried a private AI which allowed me to let go and speak freely. And I must say I’m quite impressed by the results.

When you upload your private journal entries and ask an LLM to reflect and look for trends to help you improve as a human, you get really good and interesting responses. Given the right context, these AIs will see right through your BS. They will spot your weaknesses — and then help you figure out a strategy for how to overcome them.

In short, the reports were right. Current AI models are really good at this stuff. They love spotting trends, which is super useful when you dump your notes for the entire last month into their context.

Now, the problem I face is that I won’t be able to give you any convincing examples. They are all too private and personal for me to share. It’s not so much that they are that embarrassing — it’s that even if I gave you an example, you wouldn’t understand the full weight of the discussion as you are not me. So giving up on my privacy to share an example is just not worth it.

Instead, all I can ask is that you give this a try. Find a private AI and start using it for personal development. I promise you won’t be disappointed.

We need more and better privacy

One takeaway from this experiment is that current LLMs are indeed good at sifting through text and spotting trends useful for personal development.

But the more important lesson is on the importance of privacy.

The lack of privacy had a clear chilling effect on what I used AI for.

And even worse: I didn’t grasp this until I experienced privacy. Only then did I realize how much I was holding back and “self-censoring”.

Again, this is hard to explain as it is so personal. It sounds like I have all these crazy things I want to talk with LLMs about. I don’t. I just have personal, private thoughts. Like everybody else. And I prefer they stay private. They stay in my head and in my journal entries.

I have never been good at going back over my notes to reflect on them. But I now have a tool that helps me do that. A private tool. Just like my journal is a private tool to help me process my thoughts. This new AI tool is an extension of that process. A new and different way of helping me make sense of the world. But I can only do this if I have the privacy to do so.

How to use a privacy focused AI model

Unfortunately, the world of privacy focused AI isn’t all that established. But different parties are working on it. And I’m grateful for that.

If you really want to make sure your conversations are private, you have to run your models locally. For the best open source models that is unrealistic for most of us. But the models are getting better so maybe we will soon have good ones that can run on consumer hardware.

I did not run a local model as I want something more powerful than what my hardware supports. I used the Venice.ai API together with a self hosted instance of Open WebUI. I think you can also just use the web version of Venice.ai if you want a simpler setup. But there are som caveats to the Venice.ai claim of full privacy so please do your own research before uploading anything sensitive.