Anyone who thinks Auto-Regressive LLMs are getting close to human-level AI, or merely need to be scaled up to get there, *must* read this. AR-LLMs have very limited reasoning and planning abilities. This will not be fixed by making them bigger and training them on more data.