172 karma · joined June 3, 2021
GTP is not supposed to exist according to computer science and machine learning experts in 2017. They were all wrong. Same with deep dream, styleGAN. And it will be true with the next thing.
The mental gymnastics refers to the fact that people have cognitive dissonance about GTP. They have amnesia about the decades of stagnant progress in text generation. All anyone can do is point out that it’s definitely not sentient and it’s not AGI so what’s the big deal? All of a sudden the goal posts have been moved… this is the single biggest quantum leap in text generation ever, a mind-boggling level of lucidity with grammar, punctuation and more all without ever being given a single deliberate instruction from a human being. It is objectively amazing. Objectively. And people have no reaction. Because they are not emotionally ready for it. It’s cognitive dissonance, mental gymnastics, whatever you want to call it.
One time I discussed AGI with a good friend. And I gently pressured him to play through the scenario of the advent of AGI. And he made a guilty confession to me. I could tell he was embarrassed to share his opinion. He said “can’t we just unplug it?” This is a microcosm of the entire issue. It’s something a child might say. For global warming, can’t we just turn on the air conditioning? No, we can’t just unplug it.
Here is how you can understand what is happening. An AI model is just a program. A program that is written by another program.
Consider a list of programs that contains every possible program. It is infinite. Make it finite by limiting the size of the programs based on some practical consideration. It’s still a very long list.
What is in this list? Pong. StyleGAN. AGI? Almost certainly. How small might AGI be? It probably appears more than once on the list — how many are there? But let’s ask the real question here and boil this down. How many items on the list are something we will regret having discovered? A great, great many.
Every day, people are exploring the list. They are using primitive methods to sift through all these programs and find the ones that have interesting behavior. The process of program discovery is automated with things called “training algorithms.” As our computers get faster and capital allocation grows larger we mine this list more and more quickly. The problem is that we keep finding things that surprise us. And that is the core and substance of the entire issue. We have demonstrated to ourselves over and over again that our own intuition about the contents of this list is completely wrong. Despite the mind-boggling level of mental gymnastics performed in the wake of GTP-3, enough to constitute a three ring circus, the fact of the matter is that GTP-3 and soon to be 4 is not supposed to exist according to every single AI expert. Compute is only getting cheaper and capital is only getting bigger and the list is only getting shorter and it’s not going to stop surprising everyone including the “experts.” Just because we don’t know how to make something we will truly regret doesn’t mean we aren’t capable of creating it.
Once those are in cycle orbit you now have giant water cylinders that you can catch every once in a while to ride to earth/Mars. Forever. With a tiny energy budget for corrections. Just wait till one passes by and slip your vessel inside it. Man I am good.
It’s not a question of power. You could could get it spinning quickly or slowly with high or low power respectively but the question you are trying to ask it how much energy is required. The amount of energy to get the thing spinning is the same no matter what. And once you get it spinning, you’re done. It will keep spinning. I think there is a certain law that pertains to that. Something to do with fig newtons.
The amount of energy that is required is trivial compared to the other energy budgets. And it would be spun up at low power over the course of hours or days.