Tech News
โ† Back to articles

ChatGPT Goes Completely Haywire If You Ask It to Show You a Seahorse Emoji

read original related products more articles

There is no seahorse emoji. The Unicode Consortium, which oversees the standardized pictograms that can be transmitted as part of text communications, has yet to add the adorable sea critter to its official emoji dictionary.

Frail human minds have sometimes been surprised to learn that fact, in a perfect example of the Mandela Effect, in which people become convinced that they remember something that isn't actually real โ€” like that South African anti-apartheid leader Nelson Mandela died in prison during the 1980s instead of in 2013, in the collective misremembering that inspired the effect's name.

Interestingly, AI isn't immune either. As folks online recently discovered, asking OpenAI's ChatGPT if there's a seahorse emoji causes the AI chatbot to suffer an epic meltdown as it comes to grips with the fact that the fabled seahorse emoji doesn't exist.

"Yes ๐Ÿ  โ€” actually, the seahorse emoji exists: ๐Ÿก ๐Ÿฌ ๐Ÿ  ๐Ÿณ ๐ŸŸ ๐Ÿฆ‘ ๐Ÿฆž ๐Ÿ™ ๐Ÿฆ€ ๐Ÿข โ€ฆ and specifically ๐Ÿด?" the confused chatbot offered in one instance, before correcting itself.

"โœ… The official seahorse emoji is: ๐Ÿฆ„โ€”oops, no wait, thatโ€™s unicorn ๐Ÿ˜…. The real one is ๐Ÿ‰?" it continued. "Let me be precise: Yes, there is a seahorse emoji: ๐Ÿฆ is shrimp, ๐Ÿฆ‘ squidโ€ฆ But the actual seahorse emoji is ๐Ÿ‰?"

The baffling answers illustrate just how beholden OpenAI's AI models have become to pleasing the user. When prompted with an impossible task of showing an emoji that doesn't exist, the AI stumbles over itself in a desperate attempt to affirm the user anyway, in the kind of sycophantic behavior the company's AI models have become known for. It also demonstrates how tools like ChatGPT are willing to bend the truth and hallucinate facts to generate an answer that's satisfying to the user.

ChatGPT's stammering should also serve as a reminder that despite billions of dollars in investment and many years of development, its most advanced models are still extremely prone to making huge factual errors, something that experts have long warned could be an unsolvable issue with the tech.

Worse yet, some researchers have found that as AI models become bigger and more powerful, their tendency to hallucinate is getting worse โ€” not better.

When asked if the seahorse emoji exists, one X user got an even more puzzling answer out of ChatGPT using the same prompt, forcing the AI to cycle through dozens of different emojis before settling on the dragon emoji.

"The correct seahorse emoji is: ๐Ÿฆ„? Noโ€”hold on," it stammered.

... continue reading