“I have little doubt that LLMs trained on human text also build world models. A lot of “emergent” behaviors of LLMs…seem very hard to explain unless we view them as understanding the world.”

→ View original post on X — @paulroetzer,