August 25, 2023
How to Build Your LLM-You
Do you want to train a model of yourself? LLM-You?
Do you want to train a model of yourself? LLM-You?
Start building your digital twin today, and you can leverage today’s LLM models to interact with a LLM-You that knows you by way of the data that you choose to share with it. By doing so you are simultaneously preparing for the better models of the future by building the store of contextual data about your life that a truly autonomous AI agent will need to act for you or -as- you.
1. Great training data
2. Choosing a base model
3. Training and tuning the model based on your data
The most important part, by far, is step 1. We’ll help you with all of it, but the world of machine learning and AI is evolving rapidly. The best model choice today will not be the best next year. But an investment in gathering all your personal data together will last you a lifetime (maybe even beyond).
We’ll start with the easy stuff: the words you’ve already written and the data streams your life already produces.
- notes
- blog posts
- social media posts
- documents and presentations you’ve made
- location
- calendar
- biometrics like heart rate and step count anything else you’d like to collect from the wearables and smart devices in your life
What have you forgotten? Fulcra staff can help you brainstorm and take an inventory about all the data your life is already producing.
All of the data you upload to Fulcra belongs exclusively to YOU. We’re building Fulcra into the best consumer data store on the planet, handling files (like Dropbox or any other consumer cloud storage) and streaming data (like no other consumer service on Earth). Our customers control the encryption keys to their data. Want to comb through, curate, and delete your data before using it in your model? You can do so in complete privacy.
What aspects of your life aren’t digitally observable yet?
Almost no one — the “very online” — has observability over every important aspect of their own lives. For the first time, with Fulcra, you can see where you have very rich data — what you read, for instance, or what you buy — and where the data is sparse and mostly in your head: how you spend your time at home, maybe, or who your most cherished relationships are.
We can help you measure your progress towards achieving observability in every domain of your life that you care about.
We’re experts in cutting edge instrumentation. Want to add smart beacons to measure how often you open the fridge door? Wanna upload a full body MRI every six months to track changes in your body? We got you.
What if your diary asked you questions like a curious interviewer? With the basics of your life streaming in automatically, you can log your life like history’s greatest diarists with no time wasted on the quotidian.
Next you need to train your model. All the other LLM-You prototypes have been pretty disappointing. We think that’s because:
1. Their training data is too limited (we solve that by continuously improving your observability over your own life)
2. It’s really hard to shove enough relevant personal data into the token window of a Foundation Model like GPT-4 to get it to work like it really knows you and to not confabulate what it doesn’t know
With your data in one place, with a natively hosted vector store, we reduce the friction of trying and training any model, now in the future.
We can also set up a locally hosted langchain stack for querying your data to customize queries to a closed-source Foundation Model with API access, like OpenAI’s GPT-4.
Your data is out there, but it may not be forever. A strategy shift on the part of Google, LinkedIn, etc., not to mention the cost of storing data indefinitely and regulatory risk means that the current custodians of your digital history may have strong incentives to delete your past. Your best defense against this possibility is to get your data under your control as soon as possible.
Stay in the loop with exclusive content, product updates, and news:
