And so it is in the 33rd year that the answers shall finally be revealed, but they will not bring solace or understanding. You are a bunch of randomly initialized dot products with some quasiconvex optimizer that’s tricked you into believing you are you. Consciousness is an atheistic synonym for soul.

Oh we are guided by fools. To quote the chat from one of my streams: “the AI person thing is whack”

And of course you are right. How aromantic. I spent the morning reading Eragon and adventures like that will never be possible again. Never mind the part about the dragons, I mean the part about food scarcity and a struggle for survival. Even in poor countries, the fight is not with nature, it’s with zero sum human games.

From part 59 of the unabomber manifesto, a part I find myself constantly repeating.

We divide human drives into three groups:
(1) those drives that can be satisfied with minimal effort
(2) those that can be satisfied but only at the cost of serious effort
(3) those that cannot be adequately satisfied no matter how much effort one makes.

Aka, type 2 things are challenges that are at the appropriate level for humans. Most notable is the challenge of persuading others, to work for you, to sleep with you, to spend the rest of their life with you…

Getting food used to be type 2, now it’s type 1 (or type 3 in the advent of serious nuclear war). And so it shall continue. The moving of type 2 into type 1. Robot people will not need to be persuaded. Type 2 things will require great effort to be kept around. (linked story is very worth reading)

Cthulhu may swim slowly. But he only swims left. Isn’t that interesting.

Funny how this only applies post enlightment. Technology is the only thing that changes, everything else is constant. Culture is downstream of technology. And it is technology that only swims forward.

So whether it’s me, Elon, one of you, or someone who isn’t born yet: we will build the robots. There will always be some kid who thinks the hole is the greatest thing ever…

And the line of progress will not stop at human level. As it was for chess, “human level” is a meaningless point. The world doesn’t care how you feel about this fact. AlphaZero didn’t just learn to be superhuman, it learned to be superhuman without any reference to human chess; the books, the opening theory, the personalities. It didn’t just learn to play chess, it relearned all of chess culture.

And so it shall be for human culture.

So if someone is going to do it, why not you? Come work at comma or one of the other AI labs.

  • DeepMind – Games. Thinks solving progessively more complex games in simulation is the path.
  • OpenAI – Scrape the internet. Thinks scale is all we need, aka build GPT-12.
  • Tesla – Shippability. Grit? Code a complex thing and it will become superhuman.
  • comma – Real world, real data. End to end. RL. On device learning.

This is really just a pitch to me. I’ve put almost 7 years of my life into this. Self driving is solved sans a bunch of bugfixes and scaling (still several more years). Now let’s solve life.

“My bet is that I think there’s less than six key insights that need to be made. Each one of them can probably be written on the back of an envelope. We don’t know what they are, but when they’re put together in concert with GPUs at scale and the data that we all have access to, that we can make something that behaves like a human being.”

From Carmack’s Lex Fridman. Well said.