Do not ask a language model questions involving reasoning, factuality, or moral judgment. Language is a vehicle for those, but it is not the SOURCE of them. And there is an enormous amount of language in the public sphere that’s devoid of them—and that trained the models.
@Cdespinosa When I was younger I remember reading theories that the human facility of cognition is identical to our facility of language. I think we're seeing evidence now disproving that idea.
@dgoldsmith@Cdespinosa This all feels akin to the magic trick where you ask someone for a number and given one you say “look under the lamp" - and there's a number!
What people don't realise is that for any given number there's a different place with that number.
•We• are injecting the intelligence into the LLM's responses by selecting those that match our expectations of intelligence. And all the nonsense is discarded as hallucination that can be fixed.
Add comment