TikTok says it has identified the cause of AI overview errors and inconsistencies, without detailing what this was.
But generative AI tools often make things up when responding to users, summarising or generating information, and errors can range from being hilarious to potentially harmful in nature.
Google was widely mocked in 2024 after its AI Overviews results told users to eat rocks and “glue pizza”.
Apple later faced criticism after an AI tool designed to summarise notifications created false headlines for the BBC News and the New York Times apps.
The tech giant suspended the feature, saying it would improve and update it.
Since then AI development has continued, with firms claiming the tech has vastly improved in ability and accuracy, but so-called “hallucinations” persist.
However, ChatGPT-maker OpenAI recently said it identified “goblin” and “gremlin” creeping into its systems’ responses – a quirk it believes arose after a tool it trained to have a nerdy persona incentivised mentioning the creatures.
False case law or citations appearing in court filings have meanwhile prompted warnings about AI use in legal settings, external, with AI errors also reportedly causing issues for some governments, external.