AGI is here. Feel the future of GPT5
(media.kotakuinaction2.win)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (49)
sorted by:
Not hallucinating things that take five seconds for a human to prove.
So when it doesn't hallucinate, is that good enough to say that sometimes it understands and sometimes it doesn't? Or do you need there to be no possibility of it ever being wrong?
When it hallucinates about things that no human being would ever hallucinate, you can’t trust it even when it isn’t, and you can never know if it isn’t unless you check against another source. At that point, you have no reason to use it whatsoever.
There are many situations where output is self verifying, such as the solution to math or programming questions.
If it had 100% accuracy, that would be god-like super intelligence and it would have already replaced almost every non-physical job on the planet. We're definitely not there. But we're also not at 0%.
What hallucination problems have you experienced with it?
NOPE! I’ve had LLMs directly lie to me about simple math problems. I’ve personally seen failed conversions from inches to centimeters. I don’t trust a THING they say.