Why General LLMs Will Control Physical Robots | Thijs Simonian
MTS
October 10, 2026
Thijs Simonian, Stanford junior and former OpenAI and Ella Marina builder, discusses scaling his viral Discord bot to 1.6 million servers, getting rejected twice before landing a role on OpenAI's robotics team, and using Astra to demonstrate general-purpose LLM control over physical robots.
Thijs Simonian
I think robotics is actually quite similar to computer use, in the sense that we saw very early signs of computer use and it was not very good a couple months ago. That reminds me a lot of what's going on with robot control right now. You plug a robot into your computer, ask it to do something, like pick up a block and put it in a bowl, and it's slow, it moves super slow, but it does do it. And the fact that you can do this with LMS is just incredibly exciting to me, because that means that anyone is soon gonna be able to do this. Right now, you can do it with Astra, it's okay. As these publicly released models get better and better at this technology, there's gonna be so much unlock.
SPEAKER_2
Yeah.
Hello, everyone, and welcome back to MTS. Today, we are joined in person by Thijs Simonian, who's a builder, who's worked at OpenAI, Ella Marina, Discord. He's also behind some of the most viral AI demos of the year, from a robot arm painting, the Golden Gate Bridge, like this one, to the AI wardrobe thing. We actually talked about that on MTS. He's actually still a student at Stanford. He's a junior right now. He left classes just to talk to us today. So he's skipping his classes. So this better be a great conversation. And well, I think I knew you because of Treehacks. So you're also the organizer president of Treehacks.
SPEAKER_3
Yeah, I co-directed Treehacks last year.
SPEAKER_2
Co-director of Treehacks. So great resume. Hire him for your company. No, I'm just kidding. But thank you so much for joining us today on MTS.
Thijs Simonian
Yeah, thank you so much for having me.
SPEAKER_2
Yeah, no, thank you so much for joining. Actually, I've been a little bit as peeking at your X for maybe the past few years, I think originally because of Treehacks. So that's how I came along your profile. And then this last year, all of these viral AI demos, I feel like the first one that I remember was the viral clothing saying, hey, here's how I actually used Codex to design my outfits, some sick jeans that we got on. So how did that start? What was the first viral build that you did?
Thijs Simonian
Yeah, with this clothing project and stuff? No, just all of them.
SPEAKER_2
Just like the first one that you started posting about, how did it actually start going viral? What was that like?
Thijs Simonian
Yeah, I think probably the clothing one was the first time stuff actually truly went viral.
I'd been, of course, working on projects and stuff my whole life. I mean, I'm happy to talk more about those as well, like some of the earlier stuff I've worked on.
In high school, like earliest high school, I was really into building Discord bots. And for the first time, I had one of my projects really take off on Discord. And there was a moment where I was not that good at coding. You couldn't use chat to help you do stuff at this time. I think I remember the very early GBT, I don't remember exactly what it was, but where it was just like autocomplete. And my project, this project I was working on at the time was called Truth or Dare. It was like a video game bot for Discord. And it was absolutely taking off all of a sudden, growing like tens of thousands of servers per day. And I think at this point, it probably reached around 500,000 communities, but it ended up growing up to 1.6 million, whatever. Fun scale stuff, learned a lot about scaling software and companies and projects, bringing on a team. And we were doing some super early demos of like, can you generate new AI questions with this autocomplete model? And it was like, okay, it was not very good at the time. But of course, obviously now today, you can just build the whole bot and the entire project with the models.
SPEAKER_3
No, that's really cool.
SPEAKER_2
It's funny, you were 14 years old when you were actually building out these Discord bots. You had millions of years on it at the time. This is before like GPT.
SPEAKER_3
Yes.
SPEAKER_2
Like, vibe coding was even a thing. So you've clearly been a builder early on. And I think what's really interesting about talking to you is you represent almost this new wave of builders about how you can take an idea. You can actually post it also on social media, get other people to get excited about the things that you're building. Anything is possible. It can lead to job opportunities. It can lead to whatever. The whole world is your oyster. And so you're a great representative of maybe like this younger generation where just like anything is possible. What advice would you have around like taking your idea and making it actually be something in the real world?
Read the other 20 minutes of this episode here.
$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.
Not what you expected? Full refund within 14 days.
Thousands of transcripts fetched by people building searchable podcast archives
Fetch transcripts with code
The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000794614802