brian

cs + applied math @princeton

[Kyle Wong @ewveggies]

Brutal compute hypergamy

[brian @brianmuen]

When trained only on expert trajectories, world models tend to learn the behaviour of the expert instead of the true environment dynamics.

We use an algorithm that maximizes action influence of our dataset, achieving up to 85x data efficiency.

Had a great time working on this!

[Samuel M Bateman @smbml_]

🚨 Stop feeding your world model the same boring demos. Switch to free-range play data! 🚨

Our CoRL 2026 paper finds we can optimize autonomous data collection to focus on missing action consequences, achieving up to 85x the data scaling efficiency of expert data collection!🧵

[brian @brianmuen]

@pumpkinafro1 @tomatosquadron first time watching a horror movie so didn’t know how to rate

[brian @brianmuen]

@pumpkinafro1 @tomatosquadron lmao

[brian @brianmuen]

@pumpkinafro1 I read all your posts 🥲

[brian @brianmuen]

@pumpkinafro1 LMAOOOOO pls make your account public so i can repost ts

[brian @brianmuen]

@pumpkinafro1 could have made one for me!

[brian @brianmuen]

@pumpkinafro1 no

[brian @brianmuen]

@pumpkinafro1 i'm remembering this

[brian @brianmuen]

@pumpkinafro1 imma start my twitter arc

[brian @brianmuen]

@pumpkinafro1 @sonia_clopen mb bro i miss it too

[brian @brianmuen]

@pumpkinafro1 just use the video speed controller extension

[brian @brianmuen]

@pumpkinafro1 real

[brian @brianmuen]

@pumpkinafro1 nooo