TikTok has scaled back an experimental feature that automatically generates descriptions for videos, after it went spectacularly wrong. The AI mistakenly described a video of celebrity Charli D'Amelio as a collection of blueberries with toppings, much to the amusement and confusion of users.

Initially rolled out in the US and the Philippines, the AI overviews were meant to provide context or summarize videos for TikTok users. However, they quickly became a laughing stock due to their wildly inaccurate descriptions. One user's video was described as featuring 'a person repeatedly striking their head with a rubber chicken,' while another saw an AI claim that a performance by ballroom dancers was about cutting through water.

The company has since restricted the feature to only suggest products similar to those in videos, but not for general descriptions. TikTok says users can still report and provide feedback on these summaries, yet some users suspect it's just a form of digital pranking.

As tech firms increasingly turn to AI to boost user engagement, such mishaps highlight the potential pitfalls of relying on imperfect algorithms. Meanwhile, other tech giants like Google and Apple have faced similar backlash for their own missteps with AI features, leading some to wonder if AI is ready for prime time or just a source of endless amusement and frustration.