Google LLC today announced new artificial intelligence models and access to AI tools during its Google I/O 2024 developer conference that will help with providing new resources for developers working on applications using AI and machine learning.

Up front and center, Google announced its newest model, Gemini 1.5 Flash, the company’s fastest model yet. It’s based on the Gemini Pro backend, but designed for scaling up high-frequency tasks – just like the superhero, it’s very good at running at high speed.

Like Gemini Pro, Flash is capable of multimodal experiences, through understanding text, images and audio, as well as outputting text, images and audio. It also has a 1 million-token context window.

Gemini 1.5 Pro will be receiving a private preview for a 2 million-token context window, double its already existing 1 million token context window, but developers will need to sign up for a waitlist to access it. Both models are now available in over 200 countries and territories, including the U.K. and Switzerland. 

Google’s lightweight open model family inspired by Gemini, Gemma, is getting a new sibling called PaliGemma. This new model is a vision-capable large language model that can be used for image labeling, image Q&A and other tasks on devices.

Gemma already comes in two sizes, 2B and 7B, making it small enough to run on devices such as smartphones and tablets, but now it’s coming in a beefier size with Gemma 27B in preview as part of Gemma 2, set to launch in June. It’s designed to be run on Nvidia Corp. graphics processing unit or a single Google tensor processing unit, and Google said it can outperform models that are twice its size.

Android development tools and AI

Android Studio, the tool that developers use to build and refine their mobile apps targeting the Android operating system, received an AI overhaul last year and now Google is continuing to refine the experience developers have with the AI integrated in this tool.

It’s now possible to provide custom prompts for Gemini in Studio to have it generate code suggestions right inside the development environment. The model can either create new code right in place or a developer can select a block of code and ask the model to transform it into a different or superior piece based on what it’s prompted to do – either simplify it or make it more complex. When Gemini is asked to complete a task, the user is shown how the code will change in a side window so it can be reviewed before accepting the suggestion.

Gemini has also been integrated into crash reports, allowing app quality insights to become part of the AI’s repertoire. Crashlytics and Android Vitals data can be fed into the model allowing it to generate insights into what happened when something went wrong during the runtime of an app, providing a conversational explanation from a crash report or a summary of an event. It can also provide a recommendation for next steps, including sample code.

Google previously announced that Android Studio integrated Gemini 1.0 Pro to answer Android development questions, generate code and explain best practices. Now, Gemini 1.5 Pro is being incorporated into Studio at no cost for all users.

Gemini Nano coming to Chrome

Beginning in the upcoming Chrome 126 version, Gemini Nano, the smallest model in the AI family designed to run locally on mobile devices and PCs, will be coming to the web browser natively on the desktop. Developers will be able to use Gemini to power their own AI features in apps.

With access to Gemini in Chrome, developers will be able to quickly construct apps, which have all the features of powerful AI capabilities without the need for reaching off the device into the cloud to get the work done.

“Our vision is that developers will be able to deliver powerful AI features to Chrome’s billions of users without having to worry about prompt engineering, fine-tuning, capacity and cost,” said Jon Dahlke, director of product management at Google Chrome. “All they will have to do is call a few high-level APIs, like translate, caption or transcribe.”

Chrome DevTools also got an upgrade with Gemini in the console with Chrome DevTools Console Insights, where the AI model will help generate explanations for errors and debugging solutions. DevTools helps provide developers with warnings and errors when a website is not working correctly. Those errors don’t often make sense and sometimes their origin is difficult to track down. Gemini, with its holistic view of a website’s code and access to a vast body of error codes and what causes them, can provide recommendations on how to fix them and even provide sample code.

Developer-first AI framework with Firebase Genkit

Google today introduced Firebase Genkit, an open-source framework built for JavaScript/TypeScript developers creating Node.js backends that will help them integrate AI-powered features into new and existing apps using LLMs.

Genkit will allow developers to quickly build and debug AI workloads, including built-in support for Google models, popular open-source frameworks and related services. When developers are ready to go into production, Genkit will also streamline the developer effort for deploying to Firebase or Google Cloud, and provide a console for monitoring the app when it’s live.

“Learning new technologies takes time and effort, especially when it comes to AI with rapidly evolving concepts and approaches,” the Firebase team said in the announcement. “We believe AI frameworks should reduce complexity, not add to it. Genkit is designed to feel familiar and intuitive with a minimum learning curve to get started.”

The tools provided will allow developers to initialize a project, kick off debug sessions, experiment with AI components, locally execute AI features and view complete traces of metadata for every model.

Developers will find that Genkit will put them on a road to building content generation AI apps, summarization capabilities for their products, text translation and even image generation from text prompts easily, Google said. 

Image: Google