I used to think that, but these large language models are advancing so fast because they are all the same kind of program now