Some collaborators and I recently released a first version of a benchmark we think highlights a critical gap in recent AI models in understanding causality in the real-world, beyond a physics focus.

Everyday environments are rich in tangible control interfaces (TCIs), like, light switches, appliance panels, and embedded GUIs, that are designed for humans and demand commonsense and physics reasoning, but also causal prediction and outcome verification in time and space (e.g., delayed heating, remote lights).

Paper: SWITCH: Benchmarking Modeling and Handling of Tangible Interfaces in Long-horizon Embodied Scenarios

Data and leaderboard in HuggingFace.

Feedback, suggestions, and collaborators are very welcome!

I'm hiring and can't find enough good CS candidates. Not in the UK though.

For a while I shared an old apartment with 3 others. Old building, 10th floor, cloth-wrapped wiring (no modern plastic/rubber/etc.), windows that didn't close, a condemned fenced-off balcony, and occasional rat visitors that could reach the 10th floor.

Dupe. Recent thread: https://news.ycombinator.com/item?id=45962654

[dupe] https://news.ycombinator.com/item?id=45809932

[dupe] https://news.ycombinator.com/item?id=45809932

Just maybe not enough.

Google has DeepMind, Microsoft has Microsoft Research, Meta has FAIR.

It’s not trivial to foster such environments, but they do still exist in different forms.

They seem more applied, with the possible exception of MSR. None of them seems like the "check in a few years later" sort of place.

My team and I are working on embodied AI. More specifically, focusing on humanoid legged robots for long horizon tasks combining navigation and manipulation/interaction.

We've recently put a paper out in arXiv (https://arxiv.org/pdf/2503.12533) and the project page is: https://beingbeyond.github.io/being-0/

BAAI is also hiring! No fully remote positions though. :-/