Summary
A report found Google's AI search overviews, using its Gemini model, are often inaccurate. Google disagreed with the findings, pointing out issues with the methods used to evaluate its AI. The report showed that inaccuracies in AI search results could lead to significant misinformation.
Key Facts
- Google's AI Gemini model powers search overviews.
- The report found 91% of AI search answers are correct.
- Google's AI processes over five trillion searches yearly.
- Inaccuracies can result in tens of millions of wrong answers.
- Google criticized the study, saying it had "serious holes."
- Researchers used SimpleQA to test the AI, which may have limitations.
- Google noted that AI inaccuracies included citing incorrect websites and being misled by manipulated information.
- Google provided examples where inconsistencies in source material contributed to errors.
