Personal AI – Your Personal Language Model
Yesterday chatGPT was down, yet again, so I browsed through the web looking for other Large Language Models to experiment with. In the process I came across personal AI. Personal AI is an app, at the moment, that runs on your laptop, but will eventually come to mobile phones and more.
The premise is simple. You tell the AI things about you, and it then returns answers based on what you told it. You can chat with it and it will give a response based on what you tell it.
If you tell it that you have an Apple Watch and that you use it everyday it will remember this. It can be quite slow and tedious to teach it by long winded conversations.
If you want to save time you can feed it documents. I played with this. I gae it three or four letters of motivation that I wrote, as well as my CV and some comments about work I did, and with time it learns about you, and can give more specific information.
It gives a personal score to the information it receives. It’s judging how personal the information that it receives or delivers is.
The idea of this AI is to act as a personal assistant and give answers that are true to your character, without you having to reply, because it has learned about you. The more you share with it, the more concise the answers will be.
what I’d Like To See
As I played with this AI I felt that giving it information is time consuming. I could share access with Google Drive but I am not sure, how, and where it stores my data. I prefer not to take that risk.
I would love to be able to feed xml and json files to this AI. I would love to give this AI the XML file from my blog, that reflects my thoughts for decades now. I would like to be able to feed it my dive log as well, so that it can answer questions about diving.
Personal Language Models and Large Language Models
Most AI is built around Large Language Models. They ingest entire books, websites, in fact entire libraries and more. They ingest information from Reddit, Twitter and plenty of other sites. They ingest the knwowledge from plenty of sources and can give general answers about almost anything. The G in GPT is for ‘General’.
The Personal Language Model is different. It is a specialist data set that knows about you, as an individual, rather than everyone. It learns about what you like, dislike and more. It learns what you want it to know about you. It can then answer, in your place, when you decide to share it.
When reading about where the data is stored in the Privacy section I came across that it is stored in the cloud using standard level encryption via [Oasis Labs], a company based in San Francisco. I would like to be able to store that data in Europe instead.
Privacy Concerns
At this moment in time I would not train this AI with data that is not public. That’s why a dive log and a blog are fine for training, but I wouldn’t give it access to Google Docs or other sources of information. I don’t trust that a data centre in the US values my data privacy highly enough. I would prefer to have that data in Europe.
The Cost
Although Personal AI is free to play with and experiment with, it has a professional teer at 40 USD per month or 400 per year. It has limited access to ChatGPT with the free version and priority access with the pro version.
The Quantified Self and Personal AI
The project is still young but I would love to see integration with xml and json import for dive logs, blogs and fitness apps. With these three sources of information AI could learn much more about us, with less effort. It has a lot of potential. To open up that potential we need it to ingest data more efficiently.
Two Profiles by default
Personal AI starts with two profiles by default, the personal, and professional profiles. You can train two personas if you wish too. One AI silo is for your professional self. In this silo you can tell it about where and when you worked in different places and it will be seperate from the personal AI. If you want to keep other information in the personal silo, but not the professional silo you can do this too.
And Finally
In the near future our devices will constantly be learning about us, and as they do they will be able to answer certain questions instead of us. As they know more, they will be more relevant. There are concerns about privacy but we can choose whether to allow AI to access everything we do, or restrict it. I won’t trust it with IM and more but I will trust it with blog posts, CVs, motivation letters and more.
For now it’s a fun toy to play with, and learn about.
