In a very recent paper even the authors of GPT state they were surprised that a few simple techniques can dramatically improve the performance of their AI, particularly when you tell it to show you the reasoning it used to come up with the answer that it gave. Some have suggested that in addition to improving performance this will also improve human ability to control an AI because then we would understand how it thinks and be sure that its goals were aligned with our own, however as this video which critiques the paper points out, there is some evidence the AI is not really telling us what it's thinking, but is instead telling us what it believes we want to hear. But either way the machine gives much better answers if you demand that it explain its reasoning process. Also, some have suggested that further AI improvement will soon run into a roadblock because GPT is already using nearly all of the training data that's available, however this paper gives hints as to why this is probably not the case.
Show Your Work: ChatGPT Performance Doubled w/ Process Rewards +Synthetic Data Event Horizon <https://www.youtube.com/watch?v=hZTZYffRsKI> John K Clark See what's on my new list at Extropolis <https://groups.google.com/g/extropolis> smw -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/CAJPayv0v41KjRXjdm6GzmQuXxCi-2u2jmTZ0hJWejWLTTDeEVQ%40mail.gmail.com.
