Not enough talk out there about goal-oriented AI systems. But I think going from auto-regressive token prediction to planning is a big part of it. As is goal exploration, discovery, extraction, and exploitation (ex: inverse reinforcement learning).

→ View original post on X — @alliekmiller