The article basically claims that LLMs are bad at politics and poker which is both not true (at least if they receive some level of reinforcement learning after sweep training)