Do you think it's bullshit, the claim that DeepSeek-R1 is basically on par with OpenAI’s o1?