2023-02-26

A Life Studying AI

I've been studying AI for most of my life - and that's saying something, 'cuz I'm not young.

That may seem ordinary to 30-year-old nerds, I suppose, because AI in some form has been around for decades.

The first time I had a chance to touch a computer - I was 7 or 8 years old. And that was just a calculator. It was a fancy one, with logarithms and exponents.

The church I went to when I was young had a mechanical adding machine to tabulate the donations from each service.

I used the calculator to play "guess the number", which gave me an intuitive understanding of binary trees before I even knew the idea of a "data structure".

Church of Science?

What would the church of science look like?

I believe in the scientific method. It is internally consistent, leads to discovery, and challenges itself. If someone said The Scientific Method was bad for society, I would want to see solid evidence.

On a weekly basis, I would like to have a reliable location for congregating with my fellow science-ists. I would like to hear some meaningful gospels, that help to explore, challenge, and restore our understanding of science-ism.

I would tithe when I attended such an event.

I would enjoy hearing about, and participating, in the church's missionary work. For example, I might volunteer to run a node in a model training network.

Swapping Data

An app that lets us tap our phones to each other to swap datasets.

It becomes popular to swap ChatGPT training sets. I have my machines, that are always grinding away at the the global feeds, doing their best to build me a realistic image of society that won't sunburn my brain with all the high-powered CAPO they pump into the streams.

They're always learning, and I have tweaked my red flag filters to flag and train on junk. So when I share my language datasets with you, I'm effectively telling you a chunk of what I see the world to be. It's influenced by what information I've been consuming, so it carries the imprint of my personality.

People who swap embeddings a lot are part of The Zeitgeist, while those who are less connected are less influential. Being influential on a large scale requires a large network of well connected friends. Being connected becomes currency.

That's very rough - I'm trying to sketch the idea of humans sharing embeddings as a way for language models to propagate, using meat-based-networking to bridge across air-gaps. I like that metaphor. :)