Less than three years ago, GPT-4 was the best model in the world, as it had been for a year, and we were all speculating whether there was some unique secret method that OpenAI had or whether GPT-4 was just as good as LLMs would ever get. (The answer to both questions was "no")