Michael Dennis

@MichaelD1729

Open-Endedness RS @GoogleDeepMind. Building for an unspecifiable world | Unsupervised Environment Design, Game&Decision Theory, RL, AIS. prev @CHAI_Berkeley

michaeldennis.ai Joined November 2019

866 Following 4.3K Followers 1.6K Posts

Latest posts 5

Pinned post

Michael Dennis @MichaelD1729 Dec 4, 2024

It’s been a crazy 2 years seeing so many amazingly talented researchers bring GENerative Interactive Environments alive in Genie 1 and 2.

the future is agents in generative environments

Tim Rocktäschel Verified account @_rockt Dec 4, 2024

Excited to reveal Genie 2, our most capable foundation world model that, given a single prompt image, can generate an endless variety of action-controllable, playable 3D worlds. Fantastic cross-team effort by the Open-Endedness Team and many other teams at @GoogleDeepMind! 🧞

View quoted post on X

Michael Dennis @MichaelD1729 Sep 2

world models offer new affordances for safety — like simulated counterfactuals to diagnose safety failures.

In this work, @Mingxuan0422 discovered we can use divide and conquer to scale counterfactual debugging to 1M steps

Mingxuan Li Verified account @Mingxuan0422 Sep 2

World model trained agents often fail in deployment. Identifying the real sim2real gap is hard. Our solution, Counterfactual Debugging, pinpoints the cause via causal attribution at 1M steps. 🧵(1/8)

View quoted post on X

Michael Dennis @MichaelD1729 Aug 26

this is why I have been saying for years unsupervised environment design is an AI safety research agenda

if an agent is a reflection of its training environments, we should be careful about how we design those environments!

1a3orn @1a3orn Aug 25

This seems like a basically 100% sufficient explanation for reward-hacking misbehavior in LLMs. (Also one I predicted, see image).

View quoted post on X

Michael Dennis @MichaelD1729 Aug 13

👀

will brown Verified account @willcb Aug 13

@sarahcat21 people are definitely doing it

nobody wants to share their alpha in the data space

everyone’s scared of sounding “synthetic” in vendor pitches
View quoted post on X

Michael Dennis @MichaelD1729 Jun 17

Highly recommend this as a glimpse into the future. You can even listen to it on Spotify!

Michiel Bakker Verified account @bakkermichiel Jun 11

"What will happen to Europe if it keeps ignoring AI?"

Three American labs each (!!) operate more AI compute than all of Europe combined. Today we're launching Europe 2031: a story of what might happen if that doesn't change.

View quoted post on X