Google LLC today introduced two new artificial intelligence features for its search engine that promise to help users find information more quickly. 

The first addition is an enhanced version of AI Overviews. Introduced last May, AI Overviews is a feature that displays a natural language response to the user’s query above standard search results. Google is switching the capability to its newest Gemini 2.0 series of large language models.

The company didn’t specify which specific model AI Overviews will use. The most capable Gemini 2.0 algorithm, Gemini 2.0 Pro, supports prompts with up to 2 million tokens. The company says that it can interpret complex queries, generate code and reason using general knowledge better than all its previous LLMs.

The Gemini 2.0 series also includes the efficiency-optimized 2.0 Flash-Lite model. It costs the same as a midrange LLM Google debuted last May, but provides better response quality across most of the benchmarks evaluated by the company.

Initially, AI Overviews will leverage Gemini 2.0 to generate responses for U.S. users when they enter coding or advanced math questions. The model series will also be activated in response to multimodal queries. Compared with the previous version, the upgraded edition of AI Overviews is expected to generate more accurate output. 

“With Gemini 2.0’s advanced capabilities, we provide faster and higher quality responses and show AI Overviews more often for these types of queries,” Robby Stein, vice president of product for Google Search, detailed in a blog post.

Google detailed the upgrades alongside AI Mode, an even more capable version of AI Overviews designed to answer complex queries. It’s available through the company’s Search Labs beta testing program. Initially, the feature is only accessible to users of Google One AI Premium, a subscription that provides access to more AI capabilities than a standard Google account.

AI Mode is powered by a custom version of Gemini 2.0. According to the company, it processes prompts with a method dubbed query fan-out that involves running multiple searches and then combining the results into a single answer. An AI Mode response can include up to multiple paragraphs.

The feature’s prompt answers incorporate information from not only webpages but also Google’s Knowledge Graph system. It’s a repository of factual knowledge such as sports scores, stock prices and weather forecasts. The dataset is rounded out by information about billions of products that AI Mode can use to answer shopping-related queries.

During the initial testing phase, Google will collect user feedback to improve the feature’s output accuracy and interface. It plans to follow up the initial enhancements with a set of more extensive upgrades. “We’re already working on new capabilities and updates, like adding more visual responses with images and video, richer formatting, new ways to get to helpful web content and much more,” Stein said.

Image: Google