Skip to content
TheChatArchiveThe Chat Archive
Geminigemini-2.5-pro· GoogleHugging Face LMArena Human Preference 140K

What does google's new AI mode work on Explain in detail

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with MiniMaxminimax-m1’s answer.
  1. User

    What does google's new AI mode work on Explain in detail

  2. Gemini Assistant

    Of course. Let's break down Google's new "AI mode" in detail.

    When people refer to Google's "new AI mode," they are primarily talking about two interconnected things:

    1. The Engine: The underlying AI model family called Gemini.
    2. The Main Application: The feature in Google Search now called AI Overviews (formerly known as Search Generative Experience or SGE).

    Here is a detailed explanation of what they are and how they work.


    Part 1: The Engine - The Gemini Model

    Think of Gemini as the powerful new brain that powers Google's AI features. It's not just one model but a family of models, each optimized for different tasks.

    • Gemini Ultra: The largest and most capable model, designed for highly complex tasks.
    • Gemini Pro: A high-performing, all-around model that balances capability and efficiency. This is the model most commonly used in products like the Gemini app and AI Overviews.
    • Gemini Nano: A small, highly efficient model designed to run directly on devices like your smartphone (e.g., for features in the Android operating system).

    What Makes Gemini Special? It's "Natively Multimodal."

    This is the most important concept to understand. Older AI models were often trained on one type of data (e.g., just text) and then had other capabilities (like image understanding) "bolted on" later.

    Gemini was built from the ground up to be natively multimodal. This means it was trained simultaneously on a massive dataset of:

    • Text (books, articles, websites)
    • Images (photos, diagrams, charts)
    • Audio (speech, sounds)
    • Video
    • Code

    Why this matters: Because it was trained on all these formats at once, Gemini can understand the relationships between them in a much more sophisticated way. It's like a person who can listen to you, look at a picture you're pointing at, and understand the context of your question all at the same time. This allows it to have more advanced reasoning, understanding, and problem-solving skills.


    Part 2: The Main Application - AI Overviews in Google Search

    This is the feature you see at the top of your Google search results for certain queries. It's the most visible use of the Gemini model.

    How It Works: A Step-by-Step Breakdown

    Let's say you search for something complex like: "What's a good 3-day workout routine for building muscle at home with just dumbbells?"

    Here's what Google's AI mode does behind the scenes:

    1. Deconstruct the Query: The system first breaks down your question into its core components:

      • Goal: Build muscle.
      • Location: At home.
      • Equipment: Only dumbbells.
      • Duration: 3-day routine.
      • Type of Answer Needed: A structured plan or routine.
    2. Intelligent Web Search: Instead of just finding pages with keywords, the AI performs a more conceptual search. It looks for high-quality, authoritative sources across the web, such as fitness blogs, health magazines, YouTube videos from certified trainers, and scientific articles on muscle hypertrophy.

    3. Synthesize and Corroborate Information: This is the key step. The AI doesn't just copy-paste from one source. It reads and understands information from multiple sources.

      • It looks for common patterns and corroborating advice (e.g., most sources recommend compound exercises like squats and presses).
      • It identifies different perspectives (e.g., some routines might focus on full-body workouts, while others use a "push/pull/legs" split).
      • It extracts the key pieces of information needed to build a coherent answer.
    4. Generate the AI Overview: Using the synthesized knowledge, the Gemini model generates a brand-new, conversational, and structured answer. For our example, it might look like this:

      • A brief introduction explaining the principles of a 3-day split.
      • Day 1: Push Day (Chest, Shoulders, Triceps)
        • Dumbbell Bench Press: 3 sets of 8-12 reps
        • Overhead Press: 3 sets of 8-12 reps
      • Day 2: Pull Day (Back, Biceps)
        • Dumbbell Rows: 3 sets of 8-12 reps
        • Bicep Curls: 3 sets of 10-15 reps
      • Day 3: Leg Day
        • Goblet Squats: 3 sets of 8-12 reps
        • Romanian Deadlifts: 3 sets of 8-12 reps
      • A concluding tip about progressive overload and rest.
    5. Provide Citations: Critically, the AI Overview includes links to the websites, articles, and videos it used to construct the answer. This allows you to verify the information, explore topics in more depth, and support the original content creators.

    What Kinds of Searches Does it Work On?

    AI Overviews are designed for complex queries where you would normally have to open 10 different tabs to piece together an answer. This includes:

    • Comparisons: "iOS vs. Android, what are the pros and cons for privacy?"
    • Planning: "Plan a week-long vegetarian meal plan on a budget."
    • How-To Instructions: "How to troubleshoot a slow Wi-Fi connection."
    • Brainstorming: "What are some unique gift ideas for someone who loves gardening?"

    Part 3: Where Else Does Google's New AI Work?

    The Gemini model is being integrated across Google's entire product suite.

    • Gemini App (formerly Bard): This is the dedicated chatbot where you can have long, creative, and complex conversations directly with the Gemini Pro model. You can ask it to write code, draft emails, summarize documents, and more.
    • Android & "Circle to Search": On newer Android phones, you can circle, highlight, or scribble on anything on your screen, and Google AI will search for it without you needing to switch apps. This is powered by Gemini.
    • Google Workspace (Docs, Gmail, Sheets): Features like "Help me write" in Gmail and Docs use Gemini to draft entire emails or documents based on a simple prompt. In Sheets, it can help you create templates and track information.
    • Google Photos: The new "Ask Photos" feature lets you search your photo library using natural language. For example, "Show me the best pictures from our trip to the beach last summer."
    • Developers (Vertex AI): Google provides developers and businesses access to Gemini models to build their own AI-powered applications.

    In summary, Google's "new AI mode" is a fundamental shift from a search engine that gives you a list of links to an answer engine that synthesizes information for you. It's all powered by the versatile Gemini model, with its most prominent feature being the AI Overviews that appear at the top of your search results.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.