How ChatGPT and Other LLMs Work—and Where They Could Go Next

1 Like

Reminds me of how eyewitness testimony in court (and human memory in general) has been discovered in recent decades to be less reliable than often thought. Memories can be falsified - I remember reading a science article where a researcher prompted passersby to recall meeting Bugs Bunny at Disneyland as a kid, and many of them recalled this well despite Bugs never having been a Disney character.

ChatGPT is probably less reliably knowledgeable than Jeopardy’s Watson, and upon asking ChatGPT if it or Watson would win in a fight, it told me something akin to “violence is never the answer”.

1 Like

Isn’t hallucination a specific kind of delusion? Not everyone with delusions is hallucinating.

1 Like