News for hackers. Stuff that matters.

A comment on Input to GPT Matters a Lot. Parent comment

Re:Input to GPT Matters a Lot by SketchySeaBeast on Monday April 03, 2023 @11:18AM (#35425153)
All of them, all the time? Assuming it does fail does it have sound reasoning, or is it nonsense?
[ Parent | on HN ]
Re:Input to GPT Matters a Lot by qntmfred on Monday April 03, 2023 @11:23AM (#35425264)
It doesn't have to be perfect. What generally matters is its performance relative to various human populations.
[ Parent | on HN ]
Re:Input to GPT Matters a Lot by ogogmad on Monday April 03, 2023 @11:25AM (#35425321)
Probably in between.

The point of my comment is that depending on how you interpret the description "predicts the next token", something that can do this well enough should be superhumanly clever. The task is "AI Complete".

Don't misinterpret me as saying that GPT is superhumanly clever. I'm just saying that something that's sufficiently good at what people think GPT is doing, should be superhuman.

[ Parent | on HN ]
Re:Input to GPT Matters a Lot by bogdan on Monday April 03, 2023 @11:22AM (#35425255)
Do humans have sound reasoning?
[ Parent | on HN ]
Re:Input to GPT Matters a Lot by tomrod on Monday April 03, 2023 @11:25AM (#35425318)
They have the capacity for it, despite rarely exercising it.
[ Parent | on HN ]
Re:Input to GPT Matters a Lot by SketchySeaBeast on Monday April 03, 2023 @11:45AM (#35425743)
Sometimes. Sometimes they'll come up with entirely novel ways to think through a problem that pushes knowledge forward. Do we have any examples of ChatGPT showing any such capacity?
[ Parent | on HN ]