Zhidx reported on August 9th that on Wednesday and Thursday of this week, at the Google I/O Connect China (2024 Google Developer Conference) held in Beijing, Google's technology experts from different fields around the world explained Google's latest developer tools and technologies.

The focus was on sharing Google's latest AI tools and technology updates, as well as product upgrades for developers going overseas in the fields of Web, Mobile, and Cloud. Google Developer X and Vice President and General Manager of Developer Relations, Jeanine Banks, highlighted three multimodal Gemini models: Gemini Nano, Gemini 1.5 Flash, and Gemini 1.5 Pro. The Gemini series models have been integrated into many development tools, including Android Studio, Chrome DevTools, and Project IDX. At the same time, Google's application development framework, Flutter, has been upgraded to version 3.24 and Dart 3.5, which can support developers in building cross-platform applications.

Google's President of Greater China, Chen Junyan, said that China is one of the largest developer markets in the world. In the past year, 25 Chinese developer teams have developed 31 games and applications, winning 50 Google Play Best of the Year awards in different regions around the world.

He also highlighted two major cases: in the digital marketing industry, NetEase Games improved the efficiency of advertising copy creation overseas with the help of Google Cloud Vertex AI platform, reducing the working time from more than a week to a few hours, and generating more than 700 customized copies per month; the international home online service platform Homary, based on Google AI's overseas advertising, improved the return on investment by 49% and the conversion efficiency by 50% by predicting potential purchase categories for users.

Advertisement

In addition, Google's Chinese developer website has launched a Google AI page, where developers can obtain relevant technical updates and information. The Chinese special topic pages of Codelab, Chrome, and Web have all been launched.

Firstly, the three Gemini models can meet different development needs, and Xiaomi SU7 is based on Flutter development.

Jeanine Banks, Vice President and General Manager of Google Developer X and Developer Relations, brought a sharing of AI tools and technology updates.

AI is fundamentally reshaping software development, allowing every developer to improve development efficiency. Google provides three different specifications of Gemini models, including Gemini Nano, Gemini 1.5 Flash, and Gemini 1.5 Pro.The article can be translated into English as follows:

Gemini Nano can run directly on the device, enabling fast response and protecting data privacy. Gemini 1.5 Flash is currently Google's fastest and most economical model, suitable for handling high-capacity tasks, and provides developers with a super long context window of 1 million Tokens. Gemini 1.5 Pro has higher answer quality and is suitable for handling complex tasks. Google has opened a context window of 2 million Tokens for all developers using Gemini 1.5 Pro.

A larger context window can improve model performance and help developers explore new application scenarios, but it will also consume a large amount of computing resources. Based on this, Google recently launched a context caching feature for Gemini 1.5 Flash and Gemini 1.5 Pro. Developers can cache the unchanging content in the prompts for instant recall later.

Developers who make overseas applications can use AI Studio to use these models for free, saving time for training and maintaining models.

The Gemma series models can meet developers' needs to fine-tune AI according to specific application scenarios. Jeanine said that the Gemma and Gemini series models are in the same vein, sharing underlying technologies and tools.

Google has officially released the Gemma 2 model, adding two models with parameter scales of 900 million and 2.7 billion. The 2.7 billion parameter model is optimized for Nvidia GPUs and can run efficiently on a single Nvidia GPU in Google Cloud and a single GPU in Vertex AI.

Jeanine said that the performance of Gemma 2 surpasses models with twice the parameter scale.

Gemini has been integrated into many development tools such as Android Studio and Chrome DevTools, which can obtain information such as application settings, performance data, logs, and source code, helping developers write, debug, and test code, generate documents, etc.

Google's open-source application development framework, Flutter, has been upgraded to version 3.24 and Dart 3.5, improving support for Platform View and Webview on macOS. The new API early preview version of Flutter GPU is built directly into the Flutter SDK, allowing developers to access the GPU with Dart code and enhance graphic rendering capabilities.Jeanine stated that China's level of activity in Flutter development ranks in the top three globally.

Google is building the User Scene package, which can directly import 3D models from different industry standards into Flutter projects.

The configuration application for Xiaomi's electric vehicle, the SU7, is built on Flutter. The Xiaomi car application provides a complete set of functions, enabling remote vehicle control and receiving real-time updates on vehicle status. Chen Zuobin, a senior R&D engineer for the Xiaomi car intelligent cockpit mobile app client, said that they estimate the efficiency of using Flutter is 60% higher than using native frameworks.

In addition, the early preview version of Android Studio Project IDX has been released, running entirely in the browser. Developers can quickly set up a workspace for Android Studio to start writing, building, and testing native Android applications. Both local Android Studio and cloud IDX workspaces can meet developers' development needs.

In terms of security, Firebase AI Monitoring can monitor the performance and behavior of functions built by developers using large language models in a production environment in real-time. This product is currently in a non-public preview phase. Google has also launched Checks AI Safety, which can be used to assess, monitor, and supervise the compliance of AI models and agents.

III. Gemini Nano has built-in Chrome desktop client, and multiple UI design tools have been released.

AI models on the device complement cloud models, allowing developers to deliver solutions with lower costs, fewer capacity limitations, and stronger data privacy protection, regardless of which model is used to build responsive AI functions. WebGPU and Web Assembly are the cornerstones of end-side Web AI.

WebGPU already supports 16-bit floating-point numbers, and Web Assembly integrates Memory 64 and JavaScript Promise. Based on this, the models built by developers can run efficiently on the vast majority of devices worldwide.Paul Kinlan, the head of the Google Chrome Developer Relations team, discussed how the video barrage platform Bilibili has successfully implemented a feature that prevents barrage from obscuring characters in the video by combining the use of Web Assembly and MediaPipe's image recognition technology. This has led to a 30% increase in session duration and a 19% increase in click-through rate.

Starting from the Chrome 126 browser version, Gemini Nano will be directly integrated into the Chrome desktop client. The "Help me write" feature can assist overseas users in writing product reviews, customer feedback, etc. To achieve this functionality, Google has fine-tuned the most efficient version of the Gemini model.

Google is also developing the Prompt API, which allows developers to use natural language to send instructions to the Gemini Nano in Chrome, without any data leaving the local device throughout the process.

Powerful AI capabilities are inseparable from high-quality UI, and Google has also introduced new features to help developers build applications.

Google has released the Speculation Rules API, which can improve site browsing speed by pre-reading and pre-rendering pages in the background. Aimed at single-page applications, Google has launched the View Transitions API, which can achieve smoother navigation.

In terms of debugging and optimizing applications, Chrome Devtools can issue warnings and prompts when errors occur on a website. At the same time, Chrome Devtools Insight can quickly analyze the problem and provide reasonable solutions based on the code.

III. Gemini Nano can operate without an internet connection, and Compose simplifies the cross-device application construction process.

Shi Jingyu, the head of the Google Android Developer Relations team, said that the AI functions implemented with Gemini Nano do not require an internet connection to complete all operations on the device.The Gemini Nano is implemented through the AICore system service, which is designed to address the complex challenge of managing the packaging and distribution of large language models. The AICore can manage the basic models on the device, allowing developers to use them directly.

Gemini Nano and AICore have now landed on Google Pixel 8 and Samsung Galaxy S24 series.

In the coming months, Google will further open up the usage rights of Gemini Nano.

This year, Google will support the cross-platform code reuse technology Kotlin Multiplatform for the business end, enabling developers to share code across web, server, and desktop platforms. Google has added Kotlin Multiplatform support to several Jetpack libraries, including DataStore, ViewModel, and Room.

The writing tool Compose has added features such as shared element transitions, which can help developers handle size changes and text animations during transitions, while also natively supporting predictive back gestures. Google has created a series of new Compose APIs to help developers build adaptive layouts that conform to design specifications.

Google has partnered with smartphone players such as Xiaomi, OPPO, and Samsung to integrate the device labs of these brands into the Android Device Streaming testing platform supported by FireBase. Developers can directly verify adaptive applications on real devices, including foldable smartphones and tablets.

Google has also facilitated and simplified difficult development tasks, such as the Crashlytics feature in Android Studio, which can quickly locate errors in the source code and provide repair suggestions based on Gemini's analytical capabilities.

Developers can choose to share or not share which data in the built-in AI privacy settings.IV. Deconstructing the Five Key Features of Google Cloud's New Platform, Vertex AI Solving Model Training, Tuning, and Deployment in One Stop

Google Cloud's Chief Advocate Richard Seroter introduced the five key features of the new cloud platform developed by Google.

Firstly, it is exceptionally powerful with unique advantages. The new cloud platform can bring new construction methods, automation, multifunctionality, and deep integration. Vertex AI can handle model training, tuning, and deployment in one stop.

Gemini 1.5 Pro can support a context window of over 2 million tokens, equivalent to 2 hours of video, 20 hours of audio, 1500 pages of documents, and 60,000 lines of code. When processing a 250,000-token lecture video from Google China, Gemini 1.5 Pro can simultaneously recognize the sound and images in the video and can automatically translate.

Secondly, it is self-contained and flexible for expansion. Google Cloud can configure service production set default values and support for open source interfaces. Vertex AI provides more than 150 models, including models released by Google itself, Anthropic, Meta, and others.

Thirdly, it breaks down barriers and facilitates cross-cloud convenience. The new cloud platform can tailor solutions for developers to meet their deployment needs on local JKE clusters or other cloud platforms, allowing users to migrate and manage resources between different cloud platforms. For example, BigQuery Omni allows users to run BigQuery on multiple mainstream cloud platforms without the need for data migration; AloyDB Omni provides an unrestricted database deployment environment.

Fourthly, it simplifies complexity and makes construction easy. The Google Cloud platform simplifies the construction and management process of cloud infrastructure through automation and intelligent default settings. Now developers can build a production-level cloud infrastructure within 45 minutes, achieving multiple functions.

Lastly, AI assistance is available for intelligent flexibility. Gemini for Google Cloud is deeply integrated into the cloud experience, including IDE plugins that provide code generation, completion, interpretation, and test generation functions; Gemini in Databases integrates AI intelligence into SQL generation, database operations, and other tasks.

Conclusion: Google uses new AI tools to help Chinese developers go global.Google has just launched a new suite of development tools with a significant impact, offering developers more powerful features and a more convenient interface. These tools help developers build, test, and deploy applications more quickly based on the capabilities of large models.

From the robust data processing capabilities of the Gemini API to the integrated development environment of Google AI Studio, and the cross-platform application of AI technologies, Google is continuously helping developers expand the boundaries of application construction. This may accelerate the speed at which domestic developers build applications, making their products more competitive in the global market.