[BidClub_]
Thread Guy · · 25 min

Will Depue: "It's Time To Be Optimistic About AI..."

Thread GuyWill Depue

AI & SoftwareTechnical
YouTube ↗
TL;DR
  • AI’s public case depends on solving everyday problems people already have. Depue contrasts $3,000 monthly rent with a hypothetical $5 rent pitch: “life can be less shitty in a very monotonous way.” He considers public frustration reasonable and sees no easy communications fix for labs promising economic upheaval while making enormous profits.
  • Depue expects AGI to supersede human labor. He regrets the carelessness of his “it’s over” thread but rejects reassurance that jobs will merely change; his analogy for human collaboration is a five-year-old helping an adult who could work a thousand times faster. His practical response is to pursue work one cares about, as he did by leaving OpenAI for independent projects.
  • He sees AI progress accelerating through better research, more compute and AI contributing to its own development. “Trillions of dollars of spend” are partly coming online, while models that struggle to book flights are also solving longstanding mathematical problems with proofs “beyond our imagination at the moment.” Current unevenness does not persuade him that progress is slowing.
  • His optimism does not assume permanent human control or settle machine consciousness. Depue thinks humans can control AI “for a time,” but questions whether that lasts 100, 1,000 or 10,000 years. He sees no good evidence that current AIs are conscious, while allowing that future systems—and gradual human–AI merging—might be.
  • His NVIDIA criticism centers on enforcing the policy already chosen. Depue praises Jensen Huang and declines to offer a firm judgment on whether export controls are desirable. He nevertheless thinks controls have been evaded to some extent: “if we’re gonna commit to doing something, we should follow through on it.”
  • Astronomy supplies his strongest concrete example of inexpensive AI expanding discovery. Depue describes an “analysis limited” field with large unexamined datasets and reports asteroid discoveries plus approximately two possible dwarf planets found by collaborator Josh Wolk. The latter remain conditional on confirmation; he describes them as roughly 700 kilometers wide and potentially among the solar system’s 50 largest objects.
  • Macro Grants targets the gap between capable independent researchers and willing donors. Depue argues that $200 of Claude tokens can enable significant discoveries, making certain grants unusually productive. His pitch is deliberately direct: “You’re looking for a dwarf planet? Here are some dollars.”
Digest · the substance, structured for research

1. AI’s public pitch has to reach the rent bill

  • Thread Guy contrasts ecstatic tech reactions to Starship with outsiders asking, “What does this do for me?” Depue considers that frustration reasonable: a bigger economy and gradually cheaper Chipotle mean little when rent remains $3,000 a month. Hypothetical $5 rent, more vacation and safer transportation make the promised benefits tangible.
  • Depue’s preferred pitch is that “life can be less shitty in a very monotonous way.” But better messaging has limits: labs are building technology that might disrupt jobs and make the economy rough in the medium term while earning enormous sums. “AI has a narrative problem,” he says, recalling someone else’s formulation without being certain who wrote it.
  • Thread Guy says AI-video Pokémon edits have been going viral on TikTok over roughly three months. Depue calls Sora “before its time” and “kinda mid,” but says he has not seen the current takeoff and may need to pay more attention.

2. AGI makes reassurance about changing jobs inadequate

  • Depue apologizes for a “careless thread” telling a professor that “it’s over,” but retains the underlying prediction. Conditional on AGI arriving, his formulation is “no more jobs.” As a researcher and software engineer, he expects “Fable 9” to “wipe the floor with me in every possible capacity,” and “Fable 10 and 12” to do that for everyone.
  • His collaboration analogy is a five-year-old helping an adult: “you can turn the wrench,” although the adult could finish a thousand times faster. That displacement coexists with benefits such as solving cancer and with massive risks. “It’s so over” and “we’re so back” express simultaneous convictions; he also likes his job and doing things in the world.
  • For immediate choices, Depue recommends pursuing genuine interests, estimating that more than 90% of people do not. Financial support and demand still matter. Frontier labs offer privileged access and insight, but much work remains ordinary product development inside large organizations: “it is just a normal tech company.” His own answer was to leave OpenAI, explore, read and build side projects.

3. Human intelligence may be a low ceiling for AI

  • Against Thread Guy’s growing indifference to model releases, Depue says progress is “actually rapidly accelerating.” He attributes this to better machine-learning understanding, “trillions of dollars of spend,” some compute coming online, and AI already contributing to progress. He thinks AI will soon perform much more AI research; the acceleration has several causes.
  • His evidence is the gap between models struggling to book flights and solving longstanding mathematics problems. He describes some proofs as “beyond our imagination at the moment.” Current progress rates need not persist: human intelligence runs on peanut-butter-and-jelly sandwiches, gets distracted, sleeps and needs bathroom breaks. Removing those constraints could let a system advance much faster.
  • On consciousness, Depue contrasts “meat computers” with “electrical computers,” while saying, “I don’t think the AIs are conscious just yet. I don’t think there’s good evidence for that.” He would update with evidence and believes consciousness is partially testable. Gradual human–AI merging might resemble conjoined twins with connected heads; he stresses this is a recent curiosity outside his expertise.
  • His long-run statement is stark: “the future is no longer ours,” followed by a qualification that humans can control AI “for a time.” He asks whether that holds for 100, 1,000 or 10,000 years. Thread Guy initially finds 10,000 years remote; Depue counters that even a century is within the lifespan of people alive today.

4. Export controls expose a gap between commitment and enforcement

  • Thread Guy raises the dispute over Chinese open-source models allegedly trained on NVIDIA Blackwells and asks whether Jensen Huang’s conduct has radicalized Depue. Depue calls Huang a “generational talent” and says AI would be nowhere without him. He does not offer a strong view on export controls that he trusts enough to share.
  • Depue nevertheless thinks export controls involving the U.S. government and, partly, NVIDIA have been evaded to some extent. His objection concerns execution regardless of whether the policy is wise: “if we’re gonna commit to doing something, we should follow through on it.” He calls the gap another example of government incompetence; Thread Guy agrees it is brutal.

5. Existing data and cheap analysis create a discovery opportunity

  • Depue’s astronomy thesis is that the field is “analysis limited”: vast datasets remain unexamined, while professionals concentrate on particular objects. Researchers with “your Claude and your Codex” might find something in the remainder. He says his group has found a “tremendous number of asteroids,” with details to follow the next week.
  • He credits collaborator Josh Wolk with approximately two possible dwarf planets in very distant orbits. Depue describes them as about half Pluto’s size and roughly 700 kilometers wide; if confirmed, they would rank among the solar system’s 50 largest objects. He is also searching for a ninth planet, but answers the probability question with “Who knows?”
  • Macro Grants connects people pursuing projects with wealthy backers seeking places to give. Depue argues $200 of Claude tokens can enable a significant discovery, improving the return on certain kinds of philanthropy. He says he is “pretty sure” the model used for mathematical results is the same one being shipped, without a separate model.
  • His practical starting point is to ask a model what amateur astronomers have recently discovered and where someone could contribute. Collaborator Luke is searching for vulcanoids, asteroids nearer the Sun. Finding a niche is the invitation; the ninth planet is more complicated. For Thread Guy’s simpler ambition—naming something—Depue recommends an asteroid.
Full transcript
Thread Guy

Yo, yo, yo. Mr. Will, how are you, man?

Will Depue

Great. How are you?

Thread Guy

I’m good, dude. It’s nice to meet you. I appreciate you being here with us.

Will Depue

Yeah, happy to be here.

Thread Guy

I don’t know if you’re familiar with us, so I’ll give you a quick intro on me and who our audience is, and then we can hit some fun topics. I’m a fan, first of all. I just watched your good interview with MTS the other day. It was a sick one. I also watched this pod you did about five days after joining OpenAI, like three or four years ago. It was—

Will Depue

Yeah, yeah, yeah.

Thread Guy

—you came on Gen Z Mafia. It was a pretty good artifact. We’re live every day, and we’re mostly focused on markets. So we’re talking about crypto, stocks, and obviously the AI trade as a through line. I think we’re around the same age, and I enjoy a little guilty pleasure, getting a little doomer on the side here. I know you claim to be an optimist, but you’re the most optimistic guy in SF, I guess, which is sort of a low bar.

Will Depue

Yeah.

Thread Guy

But you had that post the other day, which I know you got some flak for, about how over it is. I was talking about it on stream, which is why I was like, “Let’s get this guy on stream and talk for a little bit.” It’s an interesting time to be on here talking about AI and everything in between, so I appreciate you joining. How was SpaceX the other day?

Will Depue

Dude, unbelievable. I’m actually in a random Airbnb room right now in South Padre Island. So, yeah, flying back today. Texas was unbelievable. The Starship launch was insane.

Thread Guy

What was the energy like on the scene there?

Will Depue

Dude, just unbelievable. I got the chance to be out next to the Starship pad with all the employees. They’re there like, “We built that shit,” and they’re just going crazy. They’re like, “Whoa.”

Thread Guy

Was it packed?

Will Depue

It was insane, yeah.

Thread Guy

Wow. So, you know what’s interesting is that Twitter, tech Twitter, FinTwit, whatever, AI Twitter, is ecstatic about this launch, right? Everybody’s fired up, just believing in what’s going on. And then the other half is posting screen recordings of Instagram comments where everybody is basically like, “What does this do for me?” That’s sort of the overwhelming feeling: if you don’t work at a major lab, this doesn’t do absolutely anything for your life. I’m curious, as someone who used to work at OpenAI and then lives in a bubble, like me, but is probably more aware than I am of the outside noise, how do you feel about this growing angst from the outside? All this progress is being made within the bubble and on the front lines, but not that many people are exposed to it.

Will Depue

Yeah, no, I think that’s a totally reasonable concern. I don’t know. I think it’s good that people are frustrated and angry and questioning these things. I just don’t think this is a bad thing. I think it kind of sucks to be the average person a lot of the time. The thinker pitch is like, “Dude, the economy’s gonna get bigger, and then slowly, slowly, slowly, your Chipotle’s gonna get cheaper.” And you’re like, “Okay, man. My rent’s still like fucking $3,000 a month.” I don’t know. If we pitched AI as, “Dude, AI’s gonna make your rent $5 a month,” I’d be like, “Oh, shit. AGI, man, that sounds awesome.” Instead, we’re like, “Oh, it’s gonna solve cancer.” Dude, I don’t give a fuck about cancer.

Thread Guy

Yeah.

Will Depue

Not to say I don’t—cancer’s bad, yeah. But I don’t think about getting hit by a bus every day. It might happen. Versus if you tell me, “Dude, your cost of living’s gonna go down. You can afford more vacation days. You can travel. Life can be less shitty in a very monotonous way.” That sounds awesome, you know?

Thread Guy

You care about your existing problems that are right in front of you, like on the way to the grocery store.

Will Depue

Yeah.

Thread Guy

Right? I’m curious—

Will Depue

Yeah, if AI’s gonna make trains super clean, that’s awesome. That’s cool. Transportation’s safe now, whatever. Awesome.

Thread Guy

When you were at OpenAI, and even now, was there ever a moment where you were like, “This narrative framing has really gotten out of hand here. We should maybe dial this back or think about our podcast strategy for how we talk about the implications of AI”?

Will Depue

I think there’s a—I’m not sure if Rune tweeted this or who tweeted this. I think someone tweeted, “AI labs don’t have a narrative problem. I think AI has a narrative problem.” I just don’t think there’s any way to do it correctly. If you’re the company that’s building the technology that’s supposed to not only bring about a new technological age, but also might have massive social and economic effects, affect jobs, and make the economy somewhat rough in the medium term—and you’re also making unbelievable amounts of money doing it—how are you supposed to communicate that without coming off somewhat badly? You’re screwed, okay? You don’t need to do the—you can do all right at best, right? Everyone’s gonna hate you. You just gotta know that.

Thread Guy

So you just get it out of the way at the beginning and then try to fix it on the back end.

Will Depue

That sounds even worse. But I think the answer is just manage it the best you can and know that there’s no good way that this goes down well. It shouldn’t. I mean, it’s a really, really big change.

Thread Guy

So if you think AI is gonna be the biggest tech revolution ever, you obviously want to be positioned for it, and there’s probably no better way to be positioned for it than to work at a frontier lab. You left OpenAI. Are you interested in starting something? I don’t think you’re currently running a startup. You’re just running—

Will Depue

Yeah.

Thread Guy

—side projects. You have the grant thing you just launched, which is really cool. Do you feel like you can get more ahead as a solo entrepreneur, pursuing knowledge in this moment? Should you be trying to work at one of these emerging companies, or should you just be a solo entity, learning right now?

Will Depue

Yeah, I mean, listen, at the end of the day, OpenAI and Anthropic are still big tech companies. The majority of their employees are working on cool stuff, but still, like, “Let’s go fix the product in this way,” or “Let’s build this new ChatGPT health feature,” or whatever. I think it’s gonna be fun. Seeing from the inside is awesome. You get a lot of privileged access to understand what’s going on and why, and to talk to people. But beyond that, when it comes to doing the thing you wanna do in the world, dude, working at a big company sucks. Everybody knows this. Do you wanna work at a company with 10,000 people? If that’s your thing, go for it. But it comes with some drawbacks. In many ways, it is just a normal tech company, depending on what you work on.

Thread Guy

I’ve never had a corporate position before at a large company.

Will Depue

It’s both like—I would say this: if you’ve never done it, it’s just kinda sick to be corpo guy. You’re just like, “Man—”

Thread Guy

Yeah.

Will Depue

—I’m so— I mean, I’ve never been a true corpo guy. I’ve had a couple crappy whatever jobs, but, you know.

Thread Guy

By the way, is any part of you thinking to yourself that, as AI video stuff has really taken over the last couple of months, Sora was a little bit before its time?

Will Depue

I think Sora inherently was before its time. I can’t think of another video moment. I mean, also, Sora was kinda mid. We just had, like, “Damn, it kinda worked.”

Thread Guy

I mean, it was kinda mid, but—

Will Depue

For the time, yeah.

Thread Guy

—it’s happening right now, though. I feel like it’s happening right now with the AI video stuff.

Will Depue

You know what? I haven’t really seen it take off. Maybe I should be paying more attention. It just doesn’t come on my feed.

Thread Guy

There’s this crazy trend right now, really on TikTok, of these—

Will Depue

Mm.

Thread Guy

—really cool—also, there are all these Pokémon edits. AI video is going really viral for the last three months, I think. It’s really boomed.

Will Depue

Yeah.

Thread Guy

It’s really boomed.

Will Depue

That’s awesome. I don’t know. I have complex thoughts about AI video, but I think, in general, cool content is sick.

Thread Guy

Okay, so the tweet that I read that made me think, “Ugh, we should talk to this guy,” which you talked about with Theo a little bit, was basically like, “It’s over.” In reply to, I think she was a professor, which is even…

Thread Guy

She’s like, “I try to make it optimistic for my students.” And you’re like, “I don’t think you understand. You’re not listening to me. It’s over.” If you do believe that to be true, and I know you say it from a sort of optimistic framing—you could take it out of context, whatever it is—but my main lens is: What do you think people should be working on right now? Especially if you’re young, you’ve done well trading markets, or you’ve built some stuff before. What kind of stuff do you think transcends, and what should people be working on or thinking about right now?

Will Depue

I don’t know. I think you should think and do whatever you’re most interested in. That sounds like mid advice, except for the fact that 90% of people—or more than 90%—don’t do what they’re actually interested in or what they believe in. Obviously, there are a lot of reasons: you have to find something that financially supports you, is interesting, and that the world likes.

Thread Guy

Yeah, yeah.

Will Depue

But I think nowadays there’s more and more return in just doing whatever you think is worth doing right now, so I can’t really answer that for everyone. For me, the answer was to quit my job, go do my own side projects, explore, and read things.

On the “it’s over” comment, listen, I’m trying to— we all understand the tension between “it’s so over” and “we’re so back.” I’m saying both at the same time. Sometimes I say one, and sometimes I say the other. In the sense that AGI arrives: great, no more jobs. Okay, that’s kind of over. Fucked.

Thread Guy

Yeah, fucked.

Will Depue

I kind of like my job. I like doing things in the real world. But also, oh, cool, cancer’s solved. And these have massive risks for the world, and they’re going to totally upheave our entire social and economic structure. So I don’t know. It’s so over, we’re so back, it’s so over.

I apologize. That was a careless thread. I think I read someone giving a false narrative, like a professor saying, “Oh, guys, in the future, jobs are going to change with AI.” And I’m like, “No, they’re not.”

Thread Guy

Yeah.

Will Depue

“You’re fucked.” Dude, I’m a researcher and a software engineer, and Fable 9 will wipe the floor with me in every possible capacity. Fable 10 and 12 will do so for everyone in the world, right? There’s no collaboration, insofar as there’s as much collaboration as your five-year-old has with you when he’s trying to help you with your work.

Thread Guy

Oh.

Will Depue

No, yeah, buddy.

Thread Guy

Yeah.

Will Depue

You can turn the wrench. “Oh, good job. I could do this a thousand times faster than you, but I’ll let you do it.” So it’s over.

Thread Guy

I think it’s important, though, to have these dramatic outbursts because they lead to good conversation. Wait, I think you’ve tweeted about transhumanism before. I’ve never really talked about this on stream. Is that something you think is possible?

Will Depue

I think it’s functionally possible. I don’t know how I feel about it.

Thread Guy

How—what are your thoughts on it?

Will Depue

What specifically? There’s a lot of transhumanism, I guess.

Thread Guy

I guess I don’t know that much about it, but human-AI merge.

Will Depue

Yeah. It gets down to—I think there are only a couple of interesting questions remaining in the world. Before, there were physical questions, like how the atom works, gravity, and light. We’ve resolved a lot of these physical questions.

Thread Guy

Yeah.

Will Depue

I think the single greatest question left is: Why are we conscious and alive? If you’re not religious, you have to ask what’s functionally happening with your brain. Even if you are religious, given that God seems to have physical explanations for everything in the world, despite his existence, you have to ask: What are the physical explanations for the things in our heads?

I think this is the final thing. Unfortunately, everything points to our brains being meat computers and AIs being electrical computers. I don’t think AIs are conscious just yet. I don’t think there’s good evidence for that. But there’s no reason why they wouldn’t be in the future, right? If you’re just a big meat computer, and I can put your big meat computer in a big electrical computer, why are you conscious and that thing isn’t? There’s no good reason.

I don’t know. If more evidence comes out, I’d update differently. I think this suggests that merging might not be that impossible. It’s kind of like conjoined twins with joined heads: you can slowly connect your brain to something else, and you just slowly come together.

Thread Guy

Do you think there’s going to be a tangible goalpost that gets crossed where you’ll decide, “Okay, this thing is conscious,” or is it not that simple?

Will Depue

It is called the hard problem of consciousness for a reason.

Thread Guy

The hard problem of consciousness. Yeah.

Will Depue

It is the hard problem. I’d say that it’s incredibly hard to test, and people are pessimistic about how much you can test it. I do think you can partially test it. Beyond that, I don’t know. I don’t want to get too far into it. It’s also just not my experience. This has just been a curiosity of mine for the last few weeks.

Thread Guy

Yeah.

Thread Guy

I’m curious. Not even that long ago, we’d take a whole stream day to cover the latest model release or whatever had just come out. Recently, especially over the last three months, it’s been like, “Okay, new Fabulon,” whatever. It’s very much “whatever” over the last three months. I understand I’m not as deep in it as people actually working on the frontier. I’m more thinking about markets and the prices of things, but it’s still an underlying narrative that drives this stuff. Do you feel like the frontier has continued to accelerate faster or slower than you envisioned it would by this point, based on your prior years of experience?

Will Depue

I’d say this: I think Anthropic had a release this week.

Thread Guy

Yeah.

Will Depue

Jacob wrote a tweet about it. It seems abundantly clear that progress isn’t decelerating, but is actually rapidly accelerating.

Thread Guy

It’s crazy.

Will Depue

I don’t think that’s for a single reason. We understand more about machine learning over time, so we can make better decisions. We have way more compute. There are trillions of dollars of spend, some of which is coming online, and that accelerates progress. I think AIs are contributing to AI progress as well.

Thread Guy

Yeah.

Will Depue

Soon we’ll have AIs doing much more AI research. A good example of why this might get really crazy soon is the math results. These models can sometimes barely do something very simple for you, like book a flight, yet they’re also solving longstanding problems in mathematics. These are not human proofs. They’re beyond our imagination at the moment. And that’s today, not in three years. We’ve already solved some of the hardest problems in mathematics, which is unbelievable.

From this frame, it’s clear that humans aren’t necessarily a very smart species. We’re pretty smart for monkeys and everything else. But in terms of designing ourselves to be maximally intelligent, it’s not that hard to build a system that’s much smarter than us. Just because the rate of progress today is X, as Dwarkesh explains really well in the post, doesn’t mean it has to stay that way. It’s quite possible that once you unhook our brains from this tiny little meat sack that only has to run on PB&J sandwiches—

Thread Guy

Yeah.

Will Depue

—all we do is go to the bathroom, get distracted, and go to sleep. Suddenly that thing takes off so fast.

Thread Guy

Is that optimistic or doom to you? Do you feel optimistic about that view, that maybe humans aren’t really that smart relative to what we could be?

Will Depue

I don’t know. How do you think about optimistic versus pessimistic? I don’t really know.

Will Depue

I will say this: the future is no longer ours—or, if it is ours, it will only be ours for a bit.

Thread Guy

Awesome.

Will Depue

Sorry, I’m not saying we can’t control AI. I think we can control AI for a time. But do you think we’ll control AI for 100 years, 1,000 years, 10,000 years? In 10,000 years, is there going to be Thread Guy saying, “Yes, we’re controlling the civilization of AI agents vastly smarter than us in every dimension, and I’m directing them to do things”?

Thread Guy

10,000 years is crazy to think about, actually. Yeah. The—

Will Depue

It’s not that far—

Thread Guy

Yeah.

Will Depue

—in the grand scheme of things.

Thread Guy

In the grand scheme of things. You could imagine the power hierarchy having a little bit of a shakeup in 10,000 years.

Will Depue

Yeah.

Thread Guy

Yeah.

Will Depue

Even 100 years, dude. Nowadays, three years has been crazy. A hundred years from now isn’t that far off. There are people alive today who were alive 100 years ago.

Thread Guy

Alive 100 years ago. A century isn’t—

Will Depue

Yeah.

Thread Guy

Yeah. It’s actually a good way to think about it. Okay, you wrote this post the other day, and I think it has like 10 million views, about open-source Chinese models being trained on NVIDIA Blackwells. There was a little back-and-forth. Dylan Patel said everybody already knew this, but “everybody” is like seven people who work on the emerging frontier.

I’m curious what you think the implications are. And, second, how do you feel as someone who was working at a frontier lab, watching this Jensen situation unfold? Are you at all radicalized by seeing how this has played out over the last couple of days?

Will Depue

Yeah. I think Jensen Huang is obviously a generational talent and an unbelievable guy who’s built one of the greatest companies in the world. AI wouldn’t be anywhere without him. I don’t have super strong personal opinions on export controls that I trust deeply enough to share.

I would just say that there’s clearly some personal misbehavior, and it’s hard to quantify what the effect should be from that. That’s not up to me. I also think export controls, both by the U.S. government and partially by NVIDIA, have been somewhat evaded, maybe for various reasons. Whether or not they’re good ideas, if we’re going to commit to doing something, we should follow through on it.

It’s pretty insane that we’ve said we’re going to export-control these chips and this advanced technology, and it hasn’t been happening. I think it’s another example of government incompetence.

Thread Guy

A pretty brutal one, too, the way it’s played out. It’s been crazy to watch.

One other thing I want to talk to you about: you put out this astronomy grant of sorts that you’re helping contribute to, which I thought was really cool. I’m going to cobble together the framing of this question, because I just heard you talk about it on MTS. You were talking to Theo about all these people who have been working on space projects, pointing a telescope into the sky and collecting data forever. Maybe they were working on one specific project, but they haven’t had the time, capacity, or capability to parse through this enormous amount of data that’s already out there.

AI can come through and brute-force the data with a lot of tokens, compile it all, and look for patterns or discoveries that nobody else had the time, capability, or bandwidth to find. It made me think about the exploits happening on the internet right now. How many massive datasets exist in the world that you could pattern-match and connect to find exploits—or, on the positive side, make discoveries—that nobody had the time, bandwidth, or capability to find?

I’m curious if you could explain it the way you did on MTS, and then maybe talk about the implications across other fields. I thought it was pretty cool to hear you talk about that.

1. Discoveries Hiding In The Data

Will Depue

Dude, no, I think that was incredibly well explained. Astronomy has been analysis-limited for a long time, not intelligence-limited or people-limited. There’s a ton of data out there that we haven’t looked through. If you sit down with Claude and Codex and go through that data looking for cool stuff, chances are you might find something.

There are also a lot of objects in the universe and in the solar system, and professional astronomers are only interested in some of the most interesting ones. If you just want to have some fun and find an asteroid nobody’s found before, it feels pretty fucking awesome. You’re like, “Holy crap, that’s—”

Thread Guy

What did you find?

Will Depue

We’ve been finding and looking for a bunch of different stuff. We found a tremendous number of asteroids, which we’ll share about next week, and other minor planets.

The most interesting thing is that my collaborator, Josh Wolk, has now found approximately two dwarf planets. These are basically Pluto-like objects, about half the size of Pluto, orbiting very far out, far outside the solar system. It’s pretty sick. You get to name them. When confirmed, they’ll be among the 50 largest objects in the solar system, if they are confirmed.

Thread Guy

Top 50?

Will Depue

Top 50.

Thread Guy

And nobody’s found them before?

Will Depue

Yeah. They’re about 700 kilometers wide. It’s pretty intense.

Thread Guy

That’s crazy that nobody’s found them before.

Will Depue

Yeah. It’s wild.

Thread Guy

Who gets to name them? Who’s in charge? Is there a naming association?

Will Depue

There are some institutions for naming, and some rules around it, but you generally have some license to name it what you’d like. For larger planets, I believe the standard is that it has to be some Greek or Roman god or deity.

If we find a ninth planet—a truly large ninth planet in the solar system—it should be Planet Dionysus. I think the Dionysian era befits the AGI era: wine and grapes and plenty.

Thread Guy

How likely do you think it is that you’ll find the ninth planet?

Will Depue

Who knows? We’re looking for it right now.

Thread Guy

Is this what you’re fully focused on right now?

Will Depue

No, I have too many things to focus on, but yes, this is one of my larger focuses.

Thread Guy

What else are you primarily focused on right now?

Will Depue

I have a variety of projects I work on day to day. I have some machine-learning research projects, as you know, and some follow-ups to the stuff I posted. I’m also working on some interesting new evaluations for AI models, doing astronomy stuff, and helping get a project called Macro Grants off the ground.

The idea is basically that I know tons of sick guys who have awesome projects and don’t have money to fund them. I’m going to connect them with the guys who want to fund them, and we’re going to give money to people doing cool stuff. You’re looking for a dwarf planet? Here are some dollars. That’s all we’re doing.

Thread Guy

I liked your take, by the way. Reiterate this take about how, historically—or maybe just for astronomy, or grants in general—you would put up a bunch of money for a grant, and then the results wouldn’t come very quickly. It would be like, “Maybe at some point down the line we’ll get some results.” But now, in the AI era, you can produce results really fast, which maybe enhances the flywheel of the grant process.

Will Depue

Exactly. It’s kind of insane that you can sit down with $200 of Claude tokens and make a significant discovery. Even right now, the same model they’re using to solve these math problems is the same model they’re shipping. There isn’t a separate model. They’re just throwing that into the main line, right? I’m pretty sure this is true.

In that case, it’s up to you, Thread Guy. You could solve a major math problem if you lock in with $200 of tokens. From this view, for certain types of grant giving and philanthropy, there’s never been a higher-return time to turn my dollars into discoveries that someone else wants to make.

Honestly, weirdly enough, there’s also just a lot of money in the world.

Thread Guy

That’s a good thing. There’s a lot of money.

Will Depue

I’ve had the chance to meet a lot of fairly rich individuals who’ve built insane companies, and they’re constantly asking, “Where can I give money to people doing cool stuff?” They don’t have cool guys running around. They’re not hanging around in SF, in random Airbnbs in Boca Chica, meeting people at the Starship launch. My job is to connect those guys with these guys.

Thread Guy

Okay, this is the last thing I want to ask you, because I know you’re finding ninth planets and shit. You’ve got lots to do. Practically, I have a maxed-out Claude subscription and a super-powerful GPU. How do you go about approaching the problem of finding the ninth planet? What do you actually do?

2. How To Find Your Own Asteroid

Will Depue

I think it’s worth everyone trying. I’d recommend that people sit down and talk to the model: “What are people doing online? What recent discoveries have amateur astronomers made? How can I contribute? What’s an interesting class of object?”

For example, my other collaborator Luke has been looking for vulcanoids, which are a certain type of asteroid nearer to the Sun. Finding one of those would be super cool. It’s a different thing; you can find your own little niche and go look for stuff. I think that’s the most exciting stuff. Obviously, people are looking for exoplanets, and people are looking for asteroids. The ninth planet gets a bit more complex. Don’t even try.

Thread Guy

Yeah, I believe the ninth planet gets a little bit more complex.

Will Depue

Only slightly more complex, though. Only slightly.

Thread Guy

I just want to name something, man. That’s all I’m after.

Will Depue

Get an asteroid, dude. You can name it.

Thread Guy

That’s all I’m after here. Dude, Will, you’re awesome, man. I know you get framed as a doomer, and you’re a self-proclaimed optimist, but when you talk, you’re inspired and fired up. I think that reflects on other people as well. I appreciate you coming on, dude. It was a pleasure to meet you.

Will Depue

My pleasure.

Thread Guy

It was great to talk to you.

Will Depue

Yep. See you guys.

Thread Guy

All right. Have a good one. Peace.

Will Depue

Peace.

Thread Guy

That guy rocks, dude. I like that guy. That guy fucking rocks. I know it was off topic, but what do you mean? You get that guy to come on and you just take him, obviously. You just take him.