1) learning from experience (although you already mentioned it in the dynamic memory part) 2) physics grounding: llm-agent that can observe and act in an environment, reason spatially, plan, etc 3) integration with robotics

→ View original post on X — @marek_rosa,