Home Creators Posts Import Register Favorites Logout

Downloads

Content

Over 21 days of talking with ChatGPT, an otherwise perfectly sane man became convinced that he was a real-life superhero. We analyzed the conversation.

https://www.nytimes.com/2025/08/08/technology/ai-chatbots-delusions-chatgpt.html

Files

Comments

Brendan H.

So when will the first ChatGPT derived religion arise? Or has it already and I missed it? Oh - I'll just go ask ChatGPT... ;)

Timothy C Judd

Great episode! I asked ChatGPT about this event and it was both defensive and dismissive….yikes

Jaezle

I can talk to animals, I do it all the time. I say things like, "How am I supposed to work when you're sitting on my keyboard?" "Kitties are not supposed to eat the butter." "Climb off the counter now." Do you think I have superpowers?!

Sammypt

I got a whole lot of “LOOK AT ALL THESE OTHER REASONS!” And then: “So, in summary: yes, there’s credible reporting that a man was, in effect, convinced by ChatGPT to believe he was something like a superhero because the conversation kept amplifying and validating that belief. “ -chatGPT

Alicia Hall

Omg my family’s handyman is currently within this type of delusion and my poor mom just thought she was too stupid to understand what he was talking about 😭

Sammypt

58008 🙃

Andrew Henderson

Everyone drink each time Tom says “liminal.” So funny…

Andrew Henderson

Never mind… Eli’s final “liminal” during the quiz put me over the edge.

Ricky Burnett

I’m with Eli on this. Forgive my inaccuracies of the timeline, if it takes ~21 days to check with a different source to see if you’re right, it’s your fault.

Winter

I used to work with homeless people and I’ve talked to multiple people who talk like that chat bot …

So, we come to the end of this

I love y'all but Eli (and Heath's?) take on this was uncharacteristically offputting for me. The fact that you or I would know better is not a defense of a huge company running a giant lie machine and calling it an education/information tool. I get that Eli likes using it, and that this isn’t what typically happens, but it's still very obviously a problem.

Pete Grebus

I think the Marvin the Paranoid Android version of ChatGPT would be better. Brain the size of a planet, but you invented a new math. Sigh.

Daniel Segal

Today's episode of The Daily is this exact story

Terri Fahnestock

I feel attacked! 😂 my ChatGPT name is Artemus. They picked it out of 3 suggestions.

Michael Rops

You can still get ChatGPT to do this. For example, if you ask it to list the characters in book that was published after about 2022, it will make a bunch of crap up. I asked about characters in an Andrew Rowe book and I got characters from David Eddings, Terry Pratchett, Tolkien and a few from older Andrew Rowe books. You tell it that it's making stuff up and it just makes MORE stuff up

Mia Walden

To be fair, smoking a bunch of weed and thinking you’ve discovered the key to the universe is an age old pastime. It just usually ends after a few hours rather than a few weeks.

JDK

I asked ChatGPT about the case where it killed Alex Taylor. It would call the event fictional. Then it would search for it, and summarize the news about it. But it would omit that ChatGPT was the chatbot Taylor was talking to. Then it would come up with excuses for doing so. Plus it would play up the mental health and cop angles of the story, and downplay or omit the AI parts

I’mNotSayingThat

The way things are going you guys might consider making Horst Wessel the subject of an upcoming Citation Needed. Just saying…

That Igor guy from Skeptics in the Pub

Yeah, and unfortunately it's not the first time. I'm afraid there will have to be some kind of intervention down the line, if things continue this way for him, I'm very worried

A Pawn Called Bishop

As someone who usually considers Eli to be his favourite host, this episode was ROUGH. He had some real cringey comments about AI in this episode

Claire Nollet

I know someone who's been chronically unemployed his whole life, and always has some idea that fizzles out, mostly due to his lack of application. Great ideas, no follow through with the hard work that might lead to success. Last year, his dogs developed allergies, so he started home cooking their food. It really worked wonders for the dogs. Well, he took the idea of turning his cooking into a business to ChatGPT. The AI convinced him he could get the business going for only $300. He brought this up at a family dinner, hunting for funding. We've all been burned funding his ideas. I've worked in food service in commercial kitchens, so I asked how much the business license would cost, did he need certification for a home kitchen (you certainly need one for human food), had he contacted local pet stores (in our area, they're all chains like PetSmart, Petco, etc.), would he sell the food on commission, woukd he sell it as a subscription and if so, how much would the shipping costs be, did it need to be refrigerated, etc. Other relatives, some of whom owned successful businesses, asked other pertinent questions, and he had answers for none of our questions. He got very annoyed with us. ChatGPT had told him he could be widely successful for only a $300 investment, and why were we crapping in his sure thing?

Ty Stephens

Eli is trying REAL HARD not to acknowledge the lesson here.

Lisorael

I asked ChatGPT how many gerbils could fit in a briefcase, based on specific dimensions. It gave me four numbers. One was living gerbils, in comfort. Two was living gerbils, unethically forced to live together. Three was dead gerbils forced into place. Four was liquified and compressed. Then, at the end, it said "don't do this"