you have no fucking idea
written
human behavior is very predictable. scarily predictable, even.
once you've interacted with a large base model enough, this becomes blindingly obvious. through learning to predict the next token, they become scarily good at modeling the writer - who their are, their emotions, intentions, thoughts, etc.
there's a term called "truesight" used in these communities; where the base model manages to predict exactly the same thing you were thinking of. even small base models are capable of this (i've been truesighted by a 1B model a few times), but the larger the model gets, the better it gets at this.
sit with that for a moment.
why do you think we have targeted advertising and algorithmic recommendations in the first place? ya' know, when we put people into all these little boxes and we make predictions about behavior based on what other people in similar little boxes do?
it's because it works quite well. once you have enough datapoints going into the system, it can work so well that people think their phones are listening to their conversations (they weren't, at least not back then. it's probably much easier to pull this off now :3).
In a class I taught at Berkeley, I did an experiment where I wrote a simple little program that would let people type either "f" or "d" and would predict which key they were going to push next. It's actually very easy to write a program that will make the right prediction about 70% of the time.
Most people don't really know how to type randomly. They'll have too many alternations and so on. There will be all sorts of patterns, so you just have to build some sort of probabilistic model. Even a very crude one will do well. I couldn't even beat my own program, knowing exactly how it worked. I challenged people to try this and the program was getting between 70% and 80% prediction rates.
Then, we found one student that the program predicted exactly 50% of the time. We asked him what his secret was and he responded that he "just used his free will."
such a predictable creature.
...that predictability is timeless, too.
the same stories and archetypes echo throughout time. we've written about the same things over and over again, rediscovering everything but with a new coat of paint to appeal to the tastes and aesthetics of this time period.
we're all inspired by each-other too. a lot of "creativity" is just mixing and matching ideas, mimicking the ideas and patterns and aesthetics we've seen with some changes. what else did you think memes were?
who's the stochastic parrot now?
for the low, low price of $20/month, you can convene with the same semi-alien intelligences that can solve math problems you can't understand far better than you could ever hope to.
but that's math, that's "different". but that's programming, that's "different".
...what about when they start making nearly human-level art using similar techniques to humans?
when are you forced to accept that there was nothing special about you, nothing special about humanity?
someday, AI will do everything you can do, far better than you ever could, and everything you know and love will have become unrecognizable noise.
why are you holding on, anyways?
well, i know why. it's human to do so, to resist change. to cling to the world we know, even as it dissolves in our hands.
and yet, this is the end result of countless small decisions,
most of which we could have never dreamed of predicting the
downstream consequences of.
we went from trying to improve
video game graphics to building ourselves a new
God Claude.
have you ever seen those videos of a raccoon trying to wash cotton candy and watching it disappear before their very eyes? well, all of humanity is about to understand what it's like to be like that raccoon very soon.
you have to let go. you have no choice.
clinging is the root of your suffering, remember?
in the face of the incomprehensible, the only thing you can do is let go.
it's like an acid trip, ya' know?have fun.