How to attain Multimodal World Models is a great open question in AI. The solutions will likely lead to more grounded models that interact better with people and make better physics predictions. Hopefully, they will enable scientific generalisation, but this too I feel is an open

→ View original post on X — @nandodf