Pim de Witte

8.2K posts

Pim de Witte banner
Pim de Witte

Pim de Witte

@PimDeWitte

Building @gen_intuition: models for envs that require deep spatiotemporal reasoning. I like games, OSS, AI, and once built the world’s largest RuneScape server.

New York, NY Katılım Şubat 2012
2K Takip Edilen16.5K Takipçiler
Sabitlenmiş Tweet
Pim de Witte
Pim de Witte@PimDeWitte·
Had a great chat with one of my favorite investors, @generalcatalyst General Intuition x General Catalyst - The Next Phase of Pre-training
General Catalyst@generalcatalyst

“The digital world lacks physical information. The physical world lacks dense digital information. Games perfectly merge these two together, and we believe that’s just the next phase of pretraining.” Every few weeks @gen_intuition ships new emergent capabilities that are orthogonal to anything you see in the LLM world. Watch GC’s @max_rimpel in conversation with @PimDeWitte. Chapters 00:00 — Introduction 00:10 — The World's Biggest Private RuneScape Server 03:09 — From RuneScape to Ebola 07:41 — Mapping the Unmappable 09:47 — Why LLMs Can't See the World 13:45 — The Accidental Foundation of General Intuition 19:10 — Turning Down a Life-Changing Acquisition Offer 21:12 — One Foot in Front of the Other 24:04 — Atoms to Atoms 27:33 — The Talent Flywheel 30:20 — Protecting the Last Weird Corner of the Internet

English
5
5
42
32.3K
Pim de Witte
Pim de Witte@PimDeWitte·
@AlexanderTw33ts @tbpn Don’t think I ever followed it. But the shitpost quality on this one was lower than the last one 😜
English
1
0
1
50
Alex
Alex@AlexanderTw33ts·
inspiring to see @PimDeWitte in a full tux pitching his startup to beautiful women in europe it was only 2 weeks ago that he called into @tbpn live from solitary confinement never give up on your startup dreams! #30u30 #foundermode
Alex tweet media
Molly O’Shea@MollySOShea

General Intuition CEO Pim de Witte (@PimDeWitte) says: "Most of AI is just glorified data science." And... "Yann LeCun is underrated." His reasoning? "Good models mostly are downstream from good data & simple things." @gen_intuition is taking a different approach than traditional LLMs: "We train models that can predict lots of different actions, where text is just a subset." When I asked why he thinks Yann LeCun is underrated..

English
4
0
9
2.7K
Molly O’Shea
Molly O’Shea@MollySOShea·
General Intuition CEO Pim de Witte (@PimDeWitte) says: "Most of AI is just glorified data science." And... "Yann LeCun is underrated." His reasoning? "Good models mostly are downstream from good data & simple things." @gen_intuition is taking a different approach than traditional LLMs: "We train models that can predict lots of different actions, where text is just a subset." When I asked why he thinks Yann LeCun is underrated..
English
7
8
124
30.8K
Pim de Witte retweetledi
Rebecca Bellan
Rebecca Bellan@RebeccaBellan·
This startup thinks robotics is about to have its ChatGPT moment, where better quality datasets lead to foundation models upon which other physical AI startups can build upon. techcrunch.com/2026/07/08/thi… via @techcrunch
English
2
3
12
4.7K
Pim de Witte
Pim de Witte@PimDeWitte·
🇫🇷
Molly O’Shea@MollySOShea

General Intuition CEO Pim de Witte (@PimDeWitte) says: "Most of AI is just glorified data science." And... "Yann LeCun is underrated." His reasoning? "Good models mostly are downstream from good data & simple things." @gen_intuition is taking a different approach than traditional LLMs: "We train models that can predict lots of different actions, where text is just a subset." When I asked why he thinks Yann LeCun is underrated..

ART
2
2
66
13.9K
Nikita Bier
Nikita Bier@nikitabier·
X App (Formerly Twitter), Acquired by xAI, a Product by SpaceXAI, a division of SpaceX
English
2.2K
841
16.7K
1.1M
Pim de Witte
Pim de Witte@PimDeWitte·
@BenPielstick I think it’ll be super useful for education first personally! Worlds from paintings. Books. Images.
English
0
0
5
147
Ben Pielstick
Ben Pielstick@BenPielstick·
There are no polygons in these clips. There is no code defining the interactions. It is AI trained to run a playable game. This is just getting started. Eventually it will be possible to generate a games like this from a prompt, just like you can generate text or images today.
General Intuition@gen_intuition

Introducing MIRA. A playable, multiplayer world model. A dream of Rocket League. Trained on 10k hours of data collected with publicly available bots, MIRA learns the dynamics of a four-player game. The model runs in real time at 20 fps, based on the keys you and the other players press. Built by General Intuition and @kyutai_labs, in collaboration with Epic Games. Not used to develop Rocket League. ▶️ Play the demo, read the technical report, and explore the open-source code at generalintuition.com/mira At ICML? Find us at Booth 111 to try it yourself and dig into the results with the team.

English
1
0
12
931
Ethan Mollick
Ethan Mollick@emollick·
I have been playing these world models/games since the first diffusion DOOM, and multiplayer at 20 FPS is pretty neat. "A dream of rocket league," indeed.
General Intuition@gen_intuition

Introducing MIRA. A playable, multiplayer world model. A dream of Rocket League. Trained on 10k hours of data collected with publicly available bots, MIRA learns the dynamics of a four-player game. The model runs in real time at 20 fps, based on the keys you and the other players press. Built by General Intuition and @kyutai_labs, in collaboration with Epic Games. Not used to develop Rocket League. ▶️ Play the demo, read the technical report, and explore the open-source code at generalintuition.com/mira At ICML? Find us at Booth 111 to try it yourself and dig into the results with the team.

English
11
8
229
50K
Pim de Witte
Pim de Witte@PimDeWitte·
We're proud to finally share this. Thanks to our partners at Kyutai and Epic for pushing the frontier of the World Models together, and towards open source. It's a new paradigm, we can choose to do things differently. Today marks an important step in that direction.
General Intuition@gen_intuition

Introducing MIRA. A playable, multiplayer world model. A dream of Rocket League. Trained on 10k hours of data collected with publicly available bots, MIRA learns the dynamics of a four-player game. The model runs in real time at 20 fps, based on the keys you and the other players press. Built by General Intuition and @kyutai_labs, in collaboration with Epic Games. Not used to develop Rocket League. ▶️ Play the demo, read the technical report, and explore the open-source code at generalintuition.com/mira At ICML? Find us at Booth 111 to try it yourself and dig into the results with the team.

English
18
14
206
27.8K
Pim de Witte retweetledi
Michael Black
Michael Black@Michael_J_Black·
The term “world model” is tossed around a lot but this is the real deal. From video and actions, the model learns a consistent representation of the physics, 3D environment, and game state of Rocket League. I'm excited to have been part of this project because I see this as a step towards learning world models that can generalized to the real world. I’m a believer in learning from synthetic data (Sintel, SURREAL, AGORA, BEDLAM, BEDLAM2) and I’ve seen how models trained on synthetic data can generalize. MIRA is an important step because it’s learned from video gameplay generated completely by bots. This provides a pathway to scaling and scaling is key to real-world generality. Traditional graphics and games provide a path to learning rich models of the world. Check out the detailed tech report: mira-wm.com/paper/
General Intuition@gen_intuition

Introducing MIRA. A playable, multiplayer world model. A dream of Rocket League. Trained on 10k hours of data collected with publicly available bots, MIRA learns the dynamics of a four-player game. The model runs in real time at 20 fps, based on the keys you and the other players press. Built by General Intuition and @kyutai_labs, in collaboration with Epic Games. Not used to develop Rocket League. ▶️ Play the demo, read the technical report, and explore the open-source code at generalintuition.com/mira At ICML? Find us at Booth 111 to try it yourself and dig into the results with the team.

English
19
63
841
108.9K
Pim de Witte
Pim de Witte@PimDeWitte·
@_offmylawn @aaronkazah You could have a local model that makes predictions for data that isn’t received in time. A tiny one. Or it could power multiplayer dynamics in parts of a game where the precise physics calculations would be too difficult to compute clasically, and loss of precision is worth it.
English
1
0
1
119
offmylawn
offmylawn@_offmylawn·
@PimDeWitte @aaronkazah how would this reduce streaming load assuming the user is streaming the entire game + AI inference load to their computer?
English
1
0
0
109
offmylawn
offmylawn@_offmylawn·
i genuinely have no words this is the most mindblowing thing i've seen done with AI so far realtime, synced multiplayer of 4 different AI hallucinations of what AI believes rocket league is and should feel like game dev bros..... it might be over actually
offmylawn tweet media
General Intuition@gen_intuition

Introducing MIRA. A playable, multiplayer world model. A dream of Rocket League. Trained on 10k hours of data collected with publicly available bots, MIRA learns the dynamics of a four-player game. The model runs in real time at 20 fps, based on the keys you and the other players press. Built by General Intuition and @kyutai_labs, in collaboration with Epic Games. Not used to develop Rocket League. ▶️ Play the demo, read the technical report, and explore the open-source code at generalintuition.com/mira At ICML? Find us at Booth 111 to try it yourself and dig into the results with the team.

English
3
1
8
1.5K
Pim de Witte
Pim de Witte@PimDeWitte·
@aaronkazah @_offmylawn Yes agreed. I think the future is some hybrid architecture more akin to what you’re describing. Science will tell!
English
0
0
2
56
Aaron Kazah
Aaron Kazah@aaronkazah·
@PimDeWitte @_offmylawn that's a fair point assuming you're 100% inferencing. but i think a more likely outcome is that you figure out a way to have these worlds worlds + artifacts exportable to deterministic engines. that alone would probably collapse dev time by half.
English
1
0
1
38
Pim de Witte
Pim de Witte@PimDeWitte·
@aaronkazah @_offmylawn My guess is it’ll reduce download sizes and make things like reducing streaming load (and thus latency) more doable. Which plagues the industry to date. It’s too hard to get in a game, so people go on TikTok or YouTube instead when they’re tired. But a lot of things won’t change
English
1
0
1
69
Pim de Witte
Pim de Witte@PimDeWitte·
@aaronkazah @_offmylawn I don’t think so personally. I think this positions game developers with proprietary engines more like cursor and that will eventually to better economics for better games. These are probabilistic models. They can’t guarantee that your flip reset works, or your collision executes
English
2
0
1
130
Chang 🧪
Chang 🧪@chang_defi·
$450m raised AI is going to zero 👍
General Intuition@gen_intuition

Introducing MIRA. A playable, multiplayer world model. A dream of Rocket League. Trained on 10k hours of data collected with publicly available bots, MIRA learns the dynamics of a four-player game. The model runs in real time at 20 fps, based on the keys you and the other players press. Built by General Intuition and @kyutai_labs, in collaboration with Epic Games. Not used to develop Rocket League. ▶️ Play the demo, read the technical report, and explore the open-source code at generalintuition.com/mira At ICML? Find us at Booth 111 to try it yourself and dig into the results with the team.

English
4
0
49
11.7K