Automate your life
A blog post on my relationship with automation and how I'll change it with AI
I automate parts of my life. Not because I have to; but because it's fun. Do you know about Linux ricing culture?
"RICE" stands for Race Inspired Cosmetic Enhancement. Think Tokyo drift. You don't really need a spoiler on your Hyundai. Well, I don't really need to have a terminal based file explorer + bspwm hotkeyed + an ergodox. But I like it and it feels cool!
So, automating my life started with the biggest part of my life - my interface to computing. Screenchads can spend ~12 hours (serious) a day interfacing with a computer. So, I automated my workflow through custom hotkeys and adjusted my system to handle simple, frequent tasks. An example: updating my website takes one keystroke.
High tech low tech makes automation easier
The simpler the processes in your life, the easier it is to build technology to automate the processes. This simplicity is twofold: in the tools that you use to get things done, and in the way you live your life. The fewer things you have to automate, the less automation you need to do! Additionally, the simpler your software, the more control you have over it.
Automating navigation in your file explorer is easier when your file explorer is bash. Reduce the amount of software packages you need. Text is easier to manipulate when it's markdown or orgmode. Keep it simple.
You'd be surprised with how little software you need.
What I'm going to automate this week
I've come across some free time, and I plan to spend it automating things. First, I've already automated writing the automation itself. I plugged in openAI's chatGPT agent into my text editor - upon a keystroke, my text editor's visible contents are sent to openAI's servers (or abstractly, a streaming instruct-code trained LLM). This has increased my capacity to output code by a lot. Meta automation!
With "high LLMs" like GPT-4; the automation itself can be general, abstract, and take meta commands. This is a fancy way of saying: a lot of things that weren't automatable, are now automatable. So in proper cursed-with-engineer experience fashion; I'm going to build the infrastructure to support it!
It's event queue generated from raw data streams.
I'm going to experiment with creating a set of "observers" that effectively constantly watch me. That data then gets transformed into an intermediary textual representation - CLIP, video understanding, and whisper + some predefined schema. Then, textual output will be massaged into the predetermined representation.
The representation will be "knowledge", but I'll experiment and find out what the most useful structure is! I might find out that it is unnecessary. This is partly motivated by Meta AI’s imagebind. What’s the primitive?
The first thing I hope to automate is "verbal note taking". As I add more "things" to automate; I hope to converge to a useful structure.
Thanks for reading!
Want to keep up?
Check out my website @ yacine.ca - my drafts & notes hit up there before getting emitted out through substack
Want to hit with a one time donation? You can give me a money upvote!
Very fascinating, thanks for sharing! Is there already a github link for this?
Very exciting - this is a really interesting concept, will you be sharing the code and process as you go?