What evidence is there of LLM's containing world models? — AI Alignment Forum