Share
TikTok has scaled back an experimental artificial intelligence feature after it generated a series of “wildly inaccurate” and absurd summaries for videos on the platform.
The “AI overviews” tool, which was recently rolled out to a limited number of users in the US and the Philippines, was designed to provide additional context or summarize the events in a video. However, the feature became the subject of widespread mockery after users shared screenshots of bizarre hallucinations.
In one prominent example, first reported by Business Insider and highlighted by the BBC, a video of the famous dancer Charli D’Amelio was described as a “collection of various blueberries with different toppings.”
Other artists and celebrities, including Shakira and Olivia Rodrigo, were also reportedly met with vague or incorrect AI-generated descriptions. On Reddit, users highlighted even more concerning errors; a performance by ballroom dancers was inexplicably summarised as “a person repeatedly striking their head with a rubber chicken.”
In separate instances, videos that featured no violence or tools were described by the AI as showing someone “striking their head with a hammer.”
The errors echo the “glue pizza” moment experienced by Google’s own AI Overviews, where the search engine’s tool suggested adding non-toxic glue to pizza sauce to keep cheese from sliding off. Following the backlash and the spread of these errors on social media, TikTok has tweaked the feature’s functionality.
According to the company, the experimental summaries will now only be used to surface information about products or items similar to those seen in the videos, rather than attempting to describe the actions or identities of the people on screen.
The shift comes as major tech firms face increasing scrutiny over the reliability of AI products deployed to boost user engagement, with many tools struggling to accurately interpret visual content.
Related Posts
Discover more from Tech Digest
Subscribe to get the latest posts sent to your email.