i feel dizzy (again) - by Judah - Be Wrong
Be Wrong
SubscribeSign in
i feel dizzy (again)<br>y’all wanted a personal essay? okay, here’s a personal essay<br>Mar 15, 2024
29
Share
I saw someone say that these are supposed to be embarrassing or they’re not worth writing. I think they have a point, honesty is often uncomfortable. Unfortunately, I am incapable of defacing myself for the sake of a few dozen paragraphs, I have appearances to maintain.<br>This one isn’t a regular “let me explain a part of the world, as I see it” post either. It’s an essay that pretends to be about technological change and its consequences, but really it’s just about me, and what I think and do and feel. This is a genre I never really want to write, let alone publish. But I ended up finishing most of it in one sitting, and that hasn’t happened in a very, very long time.<br>It also came out as more pessimistic than I really am. I should therefore clarify whole essay is an emotional reaction and should not be read as financial/life/epistemic advice.<br>Oh, there’s also a suggested reading soundtrack for it1.<br>Text within this block will maintain its original spacing when published
“Professions” as you understand them will cease to exist in the medium term future - Manav Ponnekanti, 10 minutes ago.
I started writing this because I caught myself in a mood that I’d noticed thrice before. The lazy description is “existential”, a more literal one is “woah, alright, okay, steady now. I’m dizzy.”<br>The first time it happened was when the Dall-E 2 demos dropped, the second time was after they gave GPT eyes, the third happened when I saw it turn sketches into games. And the one that led to this essay was the video generation model that has somehow solved for rudimentary physics simulation and object permanence.<br>I was three hours away from publishing, and it happened AGAIN. This time, it’s finally robots.<br>These are “merely” impressive demos of capabilities I really thought would have some kind of hard limits. These are “just” high-dimensional predictions based off of an impressive (though incomplete) model of reality. These are “simply” a bunch of clever hacks and large matrices. A shiny demo isn’t everything.<br>It is still vertiginous.<br>The vertigo might just be because my beliefs about information and reality were proved very, very wrong. Everything that I thought was “incompressible” proved to be a matter of scale, it’s more than a little ridiculous.<br>There’s only so many reminders of how uncalibrated I am that I can absorb before feeling like I’m standing on ground that I shouldn’t trust. These days, my default assumption is “you can just compress good data enough, and it will spit out a good model”, which helps with the dizziness, but doesn’t change the fact (or the feeling) that the ground is shifting.<br>Text within this block will maintain its original spacing when published<br>II
“So what do we do? I don’t want to be the guy training the models”- Telegram message from Nihal in reaction to Gemini 1.5, 16th February.
Same, I don’t care enough about GPU code or training runs. Nihal is a lawyer who does research at a legal policy and governance think-tank, he enjoys reading and thinking about legal policy and governance more than any reasonable person would. I wake up and design really simple software, and the occasional beautiful website. I write my HTML by hand, because I do this as art and not to create shareholder value. This is what I want to do, am blessed enough to sometimes get paid to do, and hope I can do it for a while longer.<br>I have apps I want to make. They’re not particularly impressive ones, almost all of them are about putting text and images onto the internet. I feel rushed, because part of me still feels like there isn’t long before they become as trivial as three-digit multiplication.<br>It’s not over, of course. It’s very far from over. We work in pure simulacrum: words and ideas. Of course we’re intimidated by the word-generator. All our friends are upper-middle class (as defined by having a college degree) and want to get jobs that keep them there. Of course we’re worried about white collar workers. We would both like to make undeserved amounts of money by just thinking about things. Of course we fear digital automation.<br>We were both overreacting, but it’s still a real reaction. You’re going to see a lot more of it soon enough.<br>Most people don’t know how far along we are yet. I avoid telling them, it feels like asking for panic. And part of me believes that their panic will be unjustified, only I know how to panic properly. Only I have enough context, only I know just enough about how things work that I can see all the correct reasons to be the correct amounts of optimistic and worried. They wouldn’t get it.<br>Text within this block will maintain its original spacing when published<br>III
“Why are they teaching us linear regression when we have computers that can talk?”– A friend, after I showed him what GPT 3.5 could do.
The truth is,...