Google Expands Gemini's Multi-Step Research Feature to 40 Languages
Google brings its AI-powered in-depth research mode to 40 additional languages, tackling the complex challenge of maintaining accuracy and grammar across diverse native sources. The tool creates customized research reports through a repetitive, multi-step search process.
Google expands Gemini’s in-depth research mode to 40 additional languages, giving Google One AI premium subscribers access to a multi-step research assistant. The tool operates by first creating a customized research plan, then repeatedly searching for and extracting relevant information before compiling everything into a final comprehensive report.
This expansion introduces significant challenges for the tech giant, particularly in locating reliable sources across different languages and summarizing that information without introducing grammatical errors. HyunJeong Choe, director of engineering for the Gemini app, admits that AI overviews in native languages like Hindi currently experience summary inaccuracies, making factual correctness a top priority for the company.
To combat these issues, Google relies on native data sources, backend Google Search grounding, and native-language fact-checks before rolling out updates. Additionally, local teams review training datasets to ensure quality from native perspectives, though recent reports highlight ongoing scrutiny regarding the human rating processes used to refine these AI responses.