>>13455497Its extraordinarily unlikely that GPT-N would produce your answer given that prompt. Even adding extra random content pre-prompt to get different responses probably would never result in GPT-N giving your answer. It would probably take adding pre-prompt content correlated with that output, which is obviously not very useful. Yet you can produce that answer for whatever reason (contrarian, attention seek, etc priors), and its interesting and creative. We have a long way to go from GPT-N to AGI afaict. Unfortunately, it seems since we don't have any good models about what human intelligence actually is that encompasses creative processes, thought, deliberate action, transient priors, etc; we won't get to AGI until we develop that, probably
