Google Expands Gemini's Research Mode to 40 Languages

Gemini's Advanced Research Capabilities Expand to More Languages

Google announced on Friday an expansion of Gemini’s advanced research mode, extending its availability to 40 additional languages.

Introducing In-Depth Research Mode

Launched earlier this month, the in-depth research mode provides Google One AI premium subscribers with a powerful, AI-driven research assistant.

This function employs a multi-stage process. It begins with the creation of a detailed research plan, followed by the identification of pertinent information.

Subsequently, the tool refines its search based on the gathered data to extract key knowledge points.

Through iterative repetition of this process, a comprehensive report is ultimately generated.

Supported Languages

Gemini now supports a diverse range of languages, including:

• Arabic
• Bengali
• Chinese
• Danish
• French
• German
• Gujarati
• Hindi
• Indonesian
• Italian
• Japanese
• Kannada
• Korean
• Malayalam
• Marathi
• Polish
• Portuguese
• Swahili
• Spanish
• Tamil
• Telugu
• Thani
• Ukrainian
• Urdu

Challenges in Multilingual Implementation

A significant challenge for Google lies in identifying credible sources within each language.

Furthermore, accurately summarizing information while preserving grammatical correctness in the native language presents a complex task.

Addressing Accuracy Concerns

HyunJeong Choe, Director of Engineering for the Gemini app, discussed the issue with TechCrunch in December.

She explained that while the model is trained on clean data from trustworthy sources, AI overviews in languages like Hindi sometimes exhibit inaccuracies in their summaries.

Google relies on native data sources and utilizes Google Search to ensure information grounding.

Rigorous evaluations and fact-checking are conducted in native languages prior to model deployment.

Focus on Factuality and Training

Choe emphasized that achieving factuality is a recognized research problem within the field of generative AI.

While the model possesses substantial pre-existing knowledge, current efforts are focused on refining its ability to utilize information effectively.

Quality Assurance and Data Review

Jules Walter, Product Lead for International Markets for the Gemini app, highlighted the company’s testing programs.

These programs incorporate quality checks from native language perspectives.

Google also generates data for model training, which is then reviewed by local teams.

Recent Guidelines for Contractors

A recent TechCrunch report detailed new guidelines issued to contractors working to improve Gemini.

These guidelines prohibit contractors from skipping prompt responses, regardless of their level of expertise.

A Google spokesperson clarified that contractors assess responses not only for content but also for style and formatting.

Stay informed with TechCrunch’s AI newsletter! Subscribe here to receive it weekly.