A summary of Google I/O 2023

At the Google I/O 2023 Developer Conference held in the early hours of this morning, if there is any difference from previous years, it should be a more vigorous All in AI .

After all, after I/O announced its focus on AI last year, OpenAI's ChatGPT "stolen the house" at the end of the year. A press conference was rushed out and Bard made a factual mistake, causing the stock price to plummet by hundreds of billions of dollars in an instant, which triggered Subsequent "expanding AI investment" and layoffs of 12,000 people.

As an AI-first company, after several months of corrections, this I/O Google continued to bring AI back, and the theme of the conference was also revised: making artificial intelligence more helpful to everyone .

AI

This time, the Google CEO focused on the daily use of AI in the Google ecosystem, such as: "Help me write" in Gmail, the new immersive map view experience, the new Magic Editor experience example, etc., the main purpose is to show the impact of artificial intelligence on every help in daily life of the individual.

At the same time, PaLM 2 and Gemini released this time are also Google's new answers to AI . Among them, PaLM 2 (Pathway Language Model) is also the core content of AI in this conference. This is Google's new large-scale language model (LLM). In the preview version of PaLM 2, it focuses on improving multilingual translation, mathematics related, programming language, reasoning and natural language generation capabilities .

At present, PaLM 2 provides four size versions, suitable for different use scenarios, from small to large: Gecko gecko, Otter otter, Bison bison and Unicorn unicorn.

Among them, Gecko can even work on the mobile terminal, and it is rumored to be fast enough to run stably on the device when offline. This is one of the biggest highlights in my personal opinion. On the latest Samsung Galaxy, it can process 16 per second token .

At present, PaLM2 has penetrated into various product lines of Google, and the actual performance in the future has yet to be experienced. The specific scenarios are as follows:

• The AI ​​one-key P-map tool Magic Editor introduced earlier: magic change, seamless movement of characters

• Crack down on AI image fraud and launch image traceability functions, such as a picture has been seen elsewhere on the Internet, or when and where a similar picture appeared

• Benchmarking Microsoft Copilot, Google Workspace office suite upgrades the Duet AI brand, that is, after adding AI, hangs a brand name, and also introduces a new function called "Sidekick", which can read, summarize and answer different Google questions. Documentation issues in the application

  

• Demonstrated the Project Tailwind project, which can be considered as an AI notebook trained on user documents, positioned as a personalized tutor or writing companion

  

• Android 14 introduces AI features: can generate wallpapers and message texts , including Magic Compose and generated wallpapers (exclusive to Pixel devices), where Magic Compose function, AI can generate reply texts based on message content

  

• Launched the AI ​​​​programming robot Studio Bot for Android developers . This tool can generate code, fix bugs, and answer questions about Android application development. Studio Bot will not collect the developer's code, but will collect the conversation between the developer and Studio Bot Data, currently only US can be used in the Canary channel .

  

• Launched AI video lip-syncing technology: change the way characters speak according to the translated language, but it is not open for use, only Google authorized partners can use it for constructive projects

  

• Launched the A3 Virtual Machines supercomputer: equipped with Nvidia H100, can provide 26 exaFlops AI performance

• Launched the GameFace project, players control PC games through facial expressions, open source address: https://github.com/google/project-gameface

• Adobe firefly, with its amazing language-generated image capabilities , will be integrated directly into Bard in the coming months. In this way, in the dialogue with Bard, you can directly design the picture.

• AI chat robot Bard, new language support, support for exporting chat records to Google Docs and Gmail, currently canceling the standby, opening the English version, introducing image generation services, improving code processing, etc., and Bard can now help generate, explain and debug Code in more than 20 programming languages.

  

• There is also a Search Lab following Bard, which is mainly used to search for complex problems. It is currently an exploratory product and has not yet been fully opened. It tries to help people better analyze a complex or somewhat vague problem.

On the whole, it can be seen that AI has penetrated into various product fields of Google, so Google has always said that it is an AI-first company. There is nothing wrong with it. Although it experienced Waterloo last time, Google, which has a large number of product advantages, may really be able to regroup. Siege cities.

In addition, Gemini belongs to the next-generation multimodal basic model, and is currently working with DeepMind. From the known information, it will be a weapon for Google in the multimodal field.

hardware

In fact, Google's hardware releases are more like software, but as usual, I/O releases corresponding hardware devices every year.

Folding screen mobile phone Pixel Fold

Google's first folding screen mobile phone, the folded size is 139.7x79.5x12.1 mm, the unfolded size is 139.7x158.8x5.8 mm, and the weight is 283g , so the overall vision may be relatively small. The horizontal inward folding screen phone, Equipped with Google's self-developed Tensor G2 processor and a Titan M2 coprocessor, the price starts at $1,799 .

Pixel Tablet

After many years, Google has returned to the tablet market. The price is 499 US dollars. I remember that the previous generation should be the Nexus 7 in 2012. However, from the press conference, there is nothing special about the Pixel Tablet. It is also processed by Tensor G2. device.

Pixel 7a

The traditional talent Pixel also followed up the Pixel 7a, which also has a Tensor G2 processor, colorful color matching, and greatly improved images, starting at $499.

Summary: mediocre.

software

There have also been some updates to the software Android and some open source projects:

• Upgrade the Find My Device service: expand the category of headphones and tablets, build a tracking network for billions of Android devices, and support tracking headphones, tablets and other product categories .

  

• Cooperate with Samsung and Qualcomm to launch an immersive XR platform based on Android

  

• Android 14 Beta 2 was released for Pixel phones, mainly to fix various problems, such as fixing the problem that prevented users from interacting with notifications in the notification bar, and Google I / O also showed the lock screen clock and shortcuts. ( Android can be regarded as a passing rhythm this time )

  

• ChromeOS Beta has added support for streaming mobile apps: Chromebooks are enabled with Android 13 and newer devices such as Pixel 4a / Mi 13, and Chromebooks now support Android mobile streaming applications.

  

• Android Automotive OS 14 announced support for multiple car displays (even rear-seat linkage), a YouTube app is coming, it will more easily support porting tablet apps to Android Automotive, and it will further push web browser and game support .

  

• At this conference, Google also released Wear OS 4, and plans to introduce cloud backup into the platform and further increase battery standby time. The developer preview version was released today.

• Flutter & Dart related open source project updates, this is purely my personal concern, see: https://blog.csdn.net/ZuoYueLiang/article/details/130619319

  

• The I/O FLIP game is one of the highlights of this conference, and it is also an open source project. For details, see: https://juejin.cn/post/7231378331139997757

• This adventure Chat is obviously not as lively as last year's I/O "metaverse" venue, and it is estimated that the budget is all on AI.

  

at last

It can be seen that in the future, Google’s investment will be in AI, which also reflects that the future AI infrastructure will be the main battlefield in the world. The person in charge of Innovation Works also said about AI:

"GPT has now completely lost the mystery of NLP technology. The NLP level of all application development teams has been forcibly aligned. Everyone can compete for products and operations, but it is particularly difficult to establish an NLP-related technical moat on the application side." .

At present, the core ecology of OpenAI's ChatGPT, GPT API, and ChatGPT Plugin has begun to take shape, and the degree of burning money in this field determines that it is a winner-take-all monopoly model. Google's main opponent in the future will also be OpenAI.

So the biggest experience of this I/O is: AI is really coming ~