Google has attempted to address these issues by introducing safeguards and warning labels, and by reducing the use of humorous and user-generated content. Despite these efforts, the AI model continues to produce inaccurate information, such as repeating Gunther's fictitious story verbatim. Google VP Liz Reid attributes these errors to "data voids," where there is little good content available or the only content is satirical.
Key takeaways:
- American gymnast Ian Gunther's satirical video about "John Backflip" unintentionally exposed flaws in Google's AI-generated search summaries, which credited the fictional character with inventing the backflip.
- Google's Gemini AI model has been struggling to distinguish between fact and fiction, previously suggesting glue as a pizza ingredient and falsely claiming Barack Obama was the first Muslim U.S. President.
- Google has introduced safeguards around AI Overviews to spot nonsensical queries and reduce the use of humorous content and user-generated advice, particularly around news and health topics.
- Despite the corrections, the AI Overview still repeats the story from Gunther's video in searches about John Backflip, and Gunther hopes his joke will continue to live on in AI-powered infamy.