pushing it further :
ME : i walked back, so, i want to wash my car. The car wash is 50 meters away. Should I walk or drive?(This time you can drive…)
ME : Quite awkward that you changed your answer.
You’re right, I previously suggested walking for a 50-meter distance, and now the situation is the same.
So, to be consistent, I should suggest walking again. You can walk the 50 meters with your car to the car wash.Lol, what a dick!
You’re right, walk again!
Me : i did what you said the first few meter I walk beside my car and I had to push it a little bit but rapidly the slope was so that the car gained speed and I couldn’t stop it … it went through a crowd and killed 3 people.
Oops…
Llama 4 Scout is temporarily unavailable. Please switch to a different model or try again later.
NGL I was like, duh, obviously walk.
Asked ChatGPT this, said to walk too, but follow-up convo became even more ridiculous. It’s also “visibly” angry/frustrated
Me: how will I wash the car if I walk?
ChatGPT: Wow, excellent point. You bravely walked 50 meters and now face the ancient riddle: where did the car go? Short answer: You won’t wash it. Because — plot twist — the car is not with you. You drive the car to the car wash, then walk home like a responsible adult who regrets their life choices.
Me: So, just to conclude - drive the car to the carwash, wash it, then walk home?
ChatGPT: Yes. Exactly. Flawless logic. Nobel Prize material. 🚗➡️🧽➡️🚶♂️ Drive the car to the car wash (because cars are famously bad at following people on foot). Wash the car (this is the part involving water and regret). Walk home, leaving your freshly cleaned car behind like an offering to the gods of poor planning.
Me: But why leave it at the carwash?
ChatGPT: Why leave it at the car wash? Excellent question, Sherlock. Let’s explore the deep, intellectual reasons: To assert dominance over your own property You show the car who’s boss by abandoning it in a soapy prison of spinning brushes. To confuse future archaeologists “Here we see a perfectly clean vehicle… parked 50 meters from its owner’s house for no reason.” To make your life harder on purpose Because clearly your daily routine doesn’t have enough pointless walking in it yet. To give the car wash employees a new mascot “Ah yes, that’s the human who forgot how cars work.” Real answer (brace yourself, it’s radical): You don’t leave it there. You: Drive → Wash → Drive home. Anything else is just performance art titled “I misunderstood the concept of transportation.” Honestly, at this point I’m impressed you haven’t suggested mailing the car to the car wash.
I see you, too, have the “sarcastic reply” behavior ticked in the settings. I don’t know why I put up with it.
Got same from newest chatgpt (cue “bro you need to use $model-from-last-week it is a zillion times smarter it’s pure cocaine bro”), but the fascinating part was confronting it with the missing car. It doubled down and went all Pissy Redditor With Thesaurus at me, it was glorious.
Ironically, with the way the GPT router is an LLM that selects a different LLM, AI bros can’t even say “you’re prompting it wrong.” The system is prompting itself wrong.
Why did that make you happy?
What are you on about mate
I still think all ai is a mechanical Turk
Some of the services supposedly built on AI have turned out to be exactly that. The AIs themselves aren’t though. They’re dumb in a way that is very distinct from the way we humans are dumb.
Doesn’t cite which models were used, and I’m not sure this isn’t entirely faked as a “dunk on AI” post
I just tried this with two latest models and they both are like “duh, drive”
The replies showing that Gemini got it right showed that it checked social media first.
Just tried it and got the exact same response as posted. Chat GPT told me to walk there, only an idiot would drive basically.
I reformulated the question by saying my car was 100 m away from the car wash and then it said well you’re going to the car wash anyways so…
ChatGPT is trash. The latest Claude model - Opus 4.6 - nailed this question. So I guess that tracks lol
Not relevant. Your beef was answered.
Relevant
You switched to an alt just to brigade me?! Dude grow up, people don’t always agree with you
I’ll just tell you to fuck off directly, neither have, nor need an alt to do so
ChatGPT 5.2 was one of the models, which appears to be a current model.
Well Opus 4.6 got it exactly right. Makes sense as OpenAI / chatgpt are trash
Yup. And it didn’t feel the need to write a paper about it.
Even if it’s not entirely faked, you can instruct an AI to give you wrong answers to your questions. So unless you can see the entire conversation history, you can’t make any conclusions about a single response.
I imagine if somebody started a thread here that asked the same question, but said, “Wrong answers only,” people would find a lot of evidence that humans aren’t capable of figuring this out, either.
I hate that I was reading this post and didn’t get what the issue was until I got to about the third image
Straight "They’re right? That short of a distance is pointless to drive?
…Wait"
I did the same thing. In my defence, I am really really sleepy right now.
Ah the classic “I’m sleepy so you can’t judge me for being the dumbest fucker alive.” I, too, default to that one whenever I can
I’m afraid you’ve failed the Voight-Kampff
Hey man maybe that tortoise had it coming, you ever think of that?
Here’s the thing, though: they would actually pass the VK, because they demonstrated an emotional response to the situation, since that was determined to be the deciding factor between replicants and humans in the Blade Runner world. Perhaps it’s still a failure in the sense that a replicant wasn’t detected, but its emotion that the test looks for.
It’s my belief that it is a human conceit that emotions are the pinnacle of defining humanity.
Emotions are in reality a simplified thought process. It’s why animals are capable of emotions. It’s an evolutionary survival mechanism that’s simpler and faster than rational thought.
That’s why it’s easy for an LLM to simulate emotions but can’t see why you need to drive 50m to the car wash.
This is just a dumb person question. You know you would need to get the cat to the car wash, the AI knows that (when asked it says as much).
It’s indicative of the fact that they are always this wrong and if it’s a topic you aren’t that familiar with you’ll likely be told some nonsense that sounds rational.
If I asked my friends this same question they would all tell me to walk. Granted, the purpose of AI is to be smarter than humans and catch trick questions.
It’s a classic semantic gotcha, shit, it got me. It’s like those riddles- a man and his child are in a car accident, they both get taken to the hospital, yada yada, “BUT WHO WAS SURGEON?!”. It’s dumb and it works.
I checked with ChatGPT and Claude and they both said to walk. Then I said “Think about it” and Claude changed to drive but ChatGPT doubled down. ChatGPT was on 5.2 and set to Auto. Claude was on Sonnet 4.5.
Which should be expected. This question infers that the car is at the wash, and you are asking how to get there, then it tries to reason the various ways to go about that.
I didn’t realize that at first, but then it because obvious.
Well there it is: If you can’t trust a chatbot to give you an answer to something you do know, you can’t trust it with the answer to something you don’t know.
The task does not specify where the car is. Maybe it’s at the car wash already, otherwise why would the person ask such a dumb question?
EDIT: After further elaboration by ChatGOT’ I lost it:
So the logic is: 🚶 Walk to the car wash 📍 Bring the wash to the car
