Tag: Google Gemini

Google announced on Friday that it is expanding Gemini’s latest in-depth research mode to support 40 additional languages. This feature, introduced earlier in December, provides users of the Google One AI Premium plan access to an AI-driven research assistant capable of creating detailed reports.

What is Gemini’s In-Depth Research Mode?

The in-depth research feature operates as a multi-step tool that simplifies complex research tasks. It begins by creating a research plan, gathers relevant information, and iteratively refines its findings through repeated searches. Finally, it synthesizes the information into a comprehensive report. This process enables users to delve deeper into topics with minimal effort.

_{>>> A2681 for Apple MacBook Air 13 M2 2022 661-26150}

Supported Languages

Gemini’s expanded language capabilities now include:
Arabic, Bengali, Chinese, Danish, French, German, Gujarati, Hindi, Indonesian, Italian, Japanese, Kannada, Korean, Malayalam, Marathi, Polish, Portuguese, Swahili, Spanish, Tamil, Telugu, Thai, Ukrainian, and Urdu.
This diverse range aims to make Gemini accessible to a global audience, enhancing its utility across cultures and regions.

Challenges in Multilingual AI Research

While Gemini’s language expansion is significant, the process of providing accurate and grammatically correct summaries in native languages remains challenging. According to HyunJeong Choe, Google’s Director of Engineering for the Gemini app, the primary hurdle lies in sourcing reliable data in various languages and ensuring the AI summarizes it accurately.

Key Challenges:

1. Factual Accuracy:
   Generative AI often struggles with factual consistency, especially when processing information in languages with limited reliable data sources.
2. Grammar and Syntax:
   Summarizing information in native languages without grammatical errors is a complex task requiring advanced linguistic understanding.
3. Native Language Biases:
   Translating or summarizing content in culturally sensitive and accurate ways demands careful training and testing.

How Google Addresses These Issues

Google has implemented several strategies to improve Gemini’s performance in multilingual contexts:

1. Data Training with Native Sources:
   The model is trained using clean, trustworthy datasets in each language, ensuring the integrity of the information.
2. Search-Grounded Responses:
   Gemini relies on Google Search for additional context and grounding, making its findings more reliable.
3. Native Evaluations:
   Before releasing updates, Google conducts evaluations and fact-checking in each target language.
4. Quality Assurance Programs:
   Local teams and native speakers review datasets and responses to maintain accuracy. Jules Walter, Gemini’s Product Lead for International Markets, highlighted the importance of testing programs that incorporate feedback from native perspectives.

_{>>>21-54348-02 for Symbol PDT6800}

Choe acknowledged that factuality—ensuring the information is correct—is an ongoing research challenge in generative AI. Although pre-trained models have a vast repository of knowledge, they need to be refined to use that information effectively. Google is continuously training Gemini to handle information more reliably.

Given up on trying to get Google Assistant to dim the lights or pull your drapes? If you’re an Android user, now’s your chance to give Gemini a spin with your smart home. 

Google has just released an extension for the Gemini app that allows Google’s LLM to take charge of a wide range of smart gadgets, from smart lights and thermostats to smart shades and TVs. 

For now, you’ll need the Android version of the Gemini app, and you’ll also need to be enrolled in Google’s Public Preview program. 

Once you’ve enabled the Google Home extension on the Gemini app (this Google support page has all the details), you’ll be able to give smart home commands to Gemini that had previously been executed by Google Assistant. 

Google explains that telling Gemini the phrase “the sun is too bright in the living room” will result in the AI assistant closing the smart window blinds in your home. The company also gives an example of using “help me clean up the kitchen” to start a robot vacuum.

There are a few limitations to note, though. It’s only available on Android phones and is restricted to English prompts.For now, only the following categories are supported (according to Google): 

• Lights & power, like lights, outlets, and switches 
• Climate control, like air conditioning units, thermostats, heaters, and fans 
• Window coverings, like curtains, blinds, and shutters 
• Media devices, like TVs and speakers 
• Other smart devices, like washers, coffee makers, and vacuums 

Google warns that Gemini’s smart-home control abilities are (for the moment, anyway) “for convenience only, not safety- or security-critical purposes.” On devices that don’t have Gemini, you’ll still be using Google Assistant.

Specifically, Gemini can’t do anything with a security device that requires a PIN, stream video feeds from security cams, or perform “some” actions on gates, locks, doors, garage doors, and other security-related devices.

You can try the Google Home extension in the Gemini app now if you’re enrolled in the early preview, and a wider release is likely to follow in the future.