Aravind Srinivas, CEO of American AI company Perplexity, said in an interview with CNBC that because DeepSeek had to find a way to get around the limitations, it actually ended up building something more efficient. They came up with a lot of smart solutions. The usual perception is that the Chinese are good at copying. But the reality is that some of the details in DeepSeek 3 are excellent, and I wouldn't even be surprised that Meta would take some of them and apply them to the Llama model. It's not that they're copying, it's that they're innovating.