TikTok Suspends Flawed AI After Absurd Comparisons

⚡

Key Takeaways

1TikTok has suspended an AI feature that compared Charli D'Amelio to blueberries, following glaring errors.

2The tool, intended to add textual summaries to videos, has been criticized for its frequent and inaccurate hallucinations.

3TikTok is now focusing on product identification in videos, after negative user feedback.

💡Why it matters — This suspension highlights the ongoing challenges AI faces in accurately interpreting multimedia content, despite advancements.

Charli D'Amelio and Blueberries: TikTok's AI Goes Awry

A recently introduced artificial intelligence feature on TikTok has made headlines by comparing Charli D'Amelio, a famous content creator, to a collection of blueberries. This feature, which added textual summaries to videos, has been paused by TikTok after generating incoherent and often absurd descriptions. For instance, a dog training video was described as a work of origami. TikTok announced that the tool will now focus on identifying products present in the videos.

An AI Feature Paused

TikTok has recently suspended a new AI feature that was in testing. This decision comes after the tool produced inaccurate textual summaries for videos featuring personalities like Charli D'Amelio, Shakira, and even segments from Saturday Night Live. These "AI previews" were intended to provide additional context for videos, recommend similar products, and explain the content. However, the tool often failed, producing descriptions as fanciful as a mushroom-induced hallucination.

Examples of Confusing AI Previews

The AI previews generated descriptions that were, to say the least, confusing. For example, a video of Charli D'Amelio simply sitting in front of a white wall was described as a "collection of blueberries with different toppings." Another video, where a dog trainer explained canine behavior, was labeled as a "complex origami art exhibition." Similarly, a video of Shakira was described as a "sequence of blue shapes," and a viral post about heartbreak from a user named Victoria was summarized as "a fascinating close-up of a small hand tracing complex patterns on a smooth surface." Finally, a video of Olivia Rodrigo was described as "a face replaced by a string of letters and numbers."

User Reactions and TikTok's Adjustments

In light of these errors, a Reddit user compared the feature to a random text generator. In response to the criticism, TikTok decided to suspend the testing of this feature. A spokesperson confirmed that the tool will now be directed towards identifying products in videos, rather than attempting to describe the entire content.

A Limited Test and Undisclosed AI Models

The AI preview tool had been in testing for several months, but only for a limited number of users in the United States and a few other markets. TikTok declined to share the AI models used for this feature. However, a description within the app indicates that the tool relied either on TikTok's proprietary AI technology or on third-party solutions.

Ongoing Challenges for AIs in Understanding Content

The hallucinatory AI summaries recall the early days of ChatGPT and other AIs that often invented facts. For example, during the launch of Google’s AI previews, the tool claimed that a dog had played in the NHL and advised adding glue to a pizza. While AIs are improving, with a 90% accuracy rate according to an analysis by Oumi, these errors demonstrate that AIs can still fail in surprising ways, despite concerns about their future dominance.