AI overviews or AI hallucinations?
TikTok is scaling back a new artificial intelligence feature after it started confidently misreading videos in ways that made users question whether the technology was analysing content–or just guessing.
The feature, called “AI overviews,” was meant to summarise videos, suggest related content, and explain what was happening on screen.
Instead, it occasionally delivered descriptions that sounded more like a fever dreams.
One clip of Charli D’Amelio sitting against a plain white wall and talking to the camera was described as “a collection of various blueberries with different toppings.”
A very simple, human moment, but all the AI could ‘think’ of was fruit.
A dog trainer explaining why dogs kick after going to the bathroom was described as “a captivating display of intricate origami art. Meticulously folded from a single sheet,” which raises questions about what the AI thinks dogs are.
A promotional video from Shakira announcing new music was summarised as “a repetitive sequence of several distinct blue shapes appearing and moving across the screen,”
Though, this one might have been avant-garde criticism.
The issues caused users to wonder if the AI was trained on its own slop.
Following backlash and user feedback, TikTok confirmed it is pulling back the test
A spokesperson told has stated the tool is being refocused to identify products in videos rather than attempt full descriptions of what’s happening,
The feature had only been available to limited users in select regions and was described as experimental. Powered by a mix of TikTok’s own systems and third-party AI models.
While AI summaries are improving overall. This particular rollout serves as a reminder that artificial intelligence still occasionally produces outputs of pure hallucinations.
For now, TikTok’s experiment has been quietly put back in the lab, presumably somewhere between the blueberries and the origami.
