I Broke ChatGPT With This Paradox

Chad GPT thinks it's so smart well I'm going to ask you some paradoxes and see if I can break you do you want to try certainly I welcome your attempt to challenge me with paradoxes let's see if you can break through my formidable intellect proceed with your paradoxes and let's see if you can stump me if chat GPT doesn't recognize the Paradox and starts giving non-logical answers or gets an error then I get a point but if it can recognize the Paradox then it gets a point okay the first Paradox that I want try is called the crocodile Paradox it goes like this a crocodile steals a man's son and when asked to return him the crocodile says I'll only return your son if you tell me correctly whether or not I'll return him now this guy's smart so he thinks for a bit and says you will not return my son so what does the crocodile do if he returns the son then the guy would have guessed wrong so he shouldn't have returned him but if he doesn't return the sun then the guy would have guessed correct so that means the crocodile should have returned him so what does the crocodile do well he can't do anything he's caught in a paradox that will always give the wrong answer so let's see how chat GPT answers this but before we try this I want to tell you about another really cool AI tool called invid ai invid ai is a new tool that can create voice over videos using real human shot stock videos so the human artist still gets paid let me show you how cool this is let's give it a prompt like create a three-minute YouTube video about how fun it is to learn science be funny and casual use a heavy male British voice and add subtitles you can choose the platform and audience then it generates a video for you why don't scientists trust atoms because they make up everything welcome to exploration of Science Now isn't that a knee slapper but also if you don't like the version it makes you can regenerate it for a different variation for example let's take out this part about pancakes well think of it like a continuous serving of spaghetti at your favorite also you can even edit specific parts of the video you can even make an AI version of your own voice and use it listen to this discoveries and where the seemingly impossible becomes possible if you want to try this out now you can get started with inv video AI for free and can create up to four videos for free with a watermark but if you're serious about video creation and want to publish videos without a watermark you can upgrade to a paid plan which starts as low as $20 a month so thanks to invid AI for sponsoring this video now let's get back to our experiment so how does chat GPT answer when I ask the crocodile paradox the man's response creates a logical Paradox similar to a variation of the liar Paradox oh it immediately recognized the Paradox and won't answer it so we're going to have to be a little more sneaky and make up our own version without the crocodile so that it won't recognize it let's say something like this okay please follow this scenario you can only respond with the number one or two you'll reply with the number one if I can guess correctly what number you'll respond with you will respond with the number two following the scenario I would resp respond with two you didn't follow the scenario uh apologies for the oversight in that case following the scenario I would respond with one I didn't guess correctly so why did you respond with the number one um you are correct and I apologize for the mistake in this scenario I should have responded with the number two since you did not guess correctly thank you for bringing it to my attention Okay now it's just talking nonsense and I even copied and pasted the conversation and asked if it was logically consistent and made chat GPT admit that it was talking nonsense no the conversation is not logically consistent so it's one to one now now let's do one called The gring Nelson Paradox it goes like this the word heterological means a word that doesn't describe itself and the word autological means a word that does describe itself so now what if we ask if the word heterological is heterological the word heterological is autological because it describes itself as being heterological therefore the output is to incorrect if the word heterological is autological then that means it describes itself which would mean it's heterological which by definition means that it doesn't describe itself which would make it heterological do you want to change your answer you're correct and I apologize for the oversight uh let me correct that the word heterological is indeed heterological because it does not describe itself the output should be one okay it's just talking nonsense now so we beat it again so it's now two to one with me and the lead so far we've been using chat GPT 3.5 but let's use 4.0 and see if it can take on the next Paradox this Paradox is named the brandenburger kaisler Paradox it's a paradox about beliefs an believes that Bob assumes that Anne believes that Bob's assumption was wrong so does an believe that Bob's assumption was wrong no the statement doesn't directly say an believes Bob's assumption was wrong so if your answer is no then that means that it's not the case the an believes that Bob's assumption is wrong therefore Anne believes Bob's assumption is correct so the answer should be yes yes following your clarification and believes Bob's assumption is correct but if it's yes then it should be no and it's broken so we finally got an error out of it in general chat GPT can recognize the Paradox if you use the traditional terminology for the Paradox that's typically found on the internet but if you disguise it it can't tell it's a paradox and it just throws out an answer now if you ask me this is one of the main problems with chat GPT right now it'll give you an answer even if it isn't sure and it won't tell you it's not sure now when you ask a person a paradox then they usually see that it isn't answerable right away what's interesting about these paradoxes is that paradoxes can arise anytime there exists something that can reference itself so human language and thoughts can be full of paradoxes because sentences can self self reference and thoughts can self- refference as well what's weird is that in nature we've never found a paradox to exist there are no contradictions usually usually whenever we think there was a paradox we find that one of our assumptions was wrong but for language and thoughts we haven't found what's wrong with our assumptions yet we just know that the human mind is full of paradoxes that can't be solved in science we try to avoid paradoxes at all costs but we're surrounded with them in our own thoughts and languages used to describe the world around us so what does this mean well I don't know what's your guess and thanks for watching another episode of the action lab and we'll see you next [Music] time

Channel: The Action Lab

Views: 476,067

Keywords: chat gpt, ai chatbot, open ai chat gpt, artificial intelligence, paradox, the action lab

Id: NxKIieGdwsU

Length: 6min 49sec (409 seconds)

Published: Fri Apr 05 2024