As of now, Google Chrome has introduced three generative AI features, with promises of more innovations involving Gemini later this year. Today’s interview sheds light on the engineering behind these developments.
Chrome engineering director Adriana Porter Felt described the initiative: “We’ve been exploring ways to integrate AI technology into the browser to simplify everyday tasks such as managing tabs, utilizing Search, inputting data in forms, and navigating webpages,” she remarked. The Chrome team was actively involved in brainstorming these enhancements.
Adriana highlighted the team’s forward-thinking approach: “We continuously think about how to make Chrome as useful as possible.”
Currently, the new features do not cover “using Search” or “reading webpages” yet. Potential future capabilities could include summarizing articles or enhancing the search experience with SGE integration. Recently, the Chrome Omnibox introduced a @Gemini shortcut for direct access to gemini.google.com.
The process of developing these LLM-powered features starts with adapting the base model to specific use cases and rigorously testing its performance across diverse user scenarios.
Google unveiled the “Organize Similar Tabs” feature in January, utilizing emojis to simplify tab management visually. A significant focus was ensuring appropriate emoji use to avoid misinterpretations. Adriana pointed out the importance of context: planning a sombre event should not trigger misplaced emojis like a skull and crossbones. Collaboration with Google’s emoji team helped select safe categories like travel and nature for emoji representations.
The theme creator, modified from its initial concept due to the complexities in generating prompts, now offers a selection of dropdown menus—Subject, Style, and Mood—which streamline the creation process and prevent misuse as a general text-to-image tool.
The “Help me write” feature initially analyzes the webpage context to tailor its assistance, whether aiding in writing a restaurant review or filling out a form, ensuring relevance and precision in its support.
