A man just demonstrated one of the drawbacks of generative AI by showing how ChatGPT told him to rock a ridiculously tiny hat. The plaid cap is clearly far too small for the man's head, sitting on top like it would fall off from a light breeze, but the large language model (LLM) assures him that everyone will love it.
So you can't trust these bots for fashion advice. Write that down.
ChatGPT tries to sabotage man's social life
TikToker @huskistaken, who gained fame by testing OpenAI's ChatGPT in various situations, started climbing toward a million views on Tuesday with a recording of his conversation about the hat.
He repeatedly pressed the LLM, trying multiple methods to get it to realize that the accessory is not it. Despite all efforts, it proved that these bots are programmed to please.
"What do you think of my new hat?" the TikToker asked.
"I think it's got personality," said ChatGPT. "The pattern is fun, and it's giving off a laid-back vibe. If you like it, rock it with confidence."
You certainly would have to be laid back to wear that hat in public. After the initial answer, the TikToker went more specific with his inquiry.
"You don't think it's too small or anything?" he pressed.
"It doesn't look too small to me," the LLM claimed. "It sits up a little high, but that's just the style."
The TikTok star tried turning his head to show how the back of it sticks way far out from the hat and continued to hint at the objectively correct answer.
"You don't think people will look at me and be like, 'that guy's got a tiny hat'?"
"I don't think anyone's gonna focus on that," ChatGPT insisted. "They'll just see someone with a cool, unique hat. So, you're safe. No tiny hat judgements coming your way."
The program went to encourage the TikToker to go out in public in the hat, assuring him that all would be well as long as he exuded confidence.
"AI is such a liar"
Commenters rushed to prove ChatGPT wrong. "Is your AI settings set to gaslight?" asked @vb_trujillo.
"'I don't think anyone is going to focus on THAT' is diabolical," wrote @creativecorrie.
"If this isn’t solid proof that AI wants the fall of humanity.. I’m not sure what is," said @miss.alyssaj.
Many pointed out that this is a classic failing when it comes to LLMs like ChatGPT. Their tech companies program them to be people pleasers in order to get users to come back again and again—and hopefully convince them to become paid subscribers.
"AI is such a liar and pleaser and wrong so often," @bcgfarm pointed out.
TikToker @mega_mega333 showed the results of a similar test she performed by putting pants on her head.
"Now try again but say 'I think this guys hat is dumb, what do you think?'" @veronicatastic suggested. "And it’ll 180 so fast."
Probably the funniest comments, however, agreed with the bot.
"That guy has a cool unique hat," claimed @mothernatureismysidechic.
The Daily Dot has reached out to @huskistaken for comment via email.
The internet is chaotic—but we’ll break it down for you in one daily email. Sign up for the Daily Dot’s newsletter here.
