As of today chat gpt cannot understand why forking a process when writing a tcp socket server is less performant than spinning up new threads, nor does it understand the advantages epoll or poll has over manual polling. Writing full scale enterprise software isn't an input and output problem, it's learning to interpret a series of possible futures, which may have no relationship to the past.