After last year's cancellation, this year's Google I/O Developer Conference will be held on May 19 at 1 a.m. Beijing time. Regarding the content covered in the opening speech, Google said: "Please pay attention to how we further fulfill our mission to organize global information and make it universally accessible and useful."
Live Video Address:
Google I/O 2021 | YouTube
Raghavan returned to the stage and started talking about augmented reality. Google brings the AR avatar of a famous athlete to search.
(Many people may be confused about the scene of this Google I/O scene, and the official directly shared a real background photo)
Next up is Google Lens, which now combines translations with "learning resources" on the web.
Raghavan shows a very complex question about preparation for mountaineering. Example: "I've climbed Mount Adams and I'm planning to climb Mount Fuji next fall, so what else should I do?" The result of the feedback was: "Mt. Fuji is about the same height, but the temperature is colder, so please prepare relevant equipment (supplemented by translatable Japanese source information)".
Prabhakar Raghavan took the stage to talk about search. The Multitasking Unified Model (MUM), while built on top of a converter architecture, is 1,000 times more powerful than BERT. MUM gained in-depth knowledge, language understanding, training in more than 75 languages, and an understanding of videos and images.
"We developed the Android Private Compute Core." Foreign media pointed out that although Google continues to emphasize privacy protection, it has not been able to respond to Apple's app tracking transparency policy on the iPhone. Fitzpatrick uses plenty of close-ups to convey Google's message: They're trying to protect users' privacy.
Being able to place a button at the top of the settings to quickly delete history is an improvement that is obviously quite welcome. Now, it's easier for users to delete content that falls on Google Search. Google Maps periodically reminds users to open location history.
The Google Pixel smartphone will usher in the "Locked Photos" section of the Albums app, and other Android smartphones will also receive pushes. These photos that are placed in the dropbox are not displayed by default in the user's photo stream.
In terms of privacy sandbox, Google clearly wants to replace third-party cookies with FLoC, although this matter has not yet gained wide acceptance in the industry. Both Differential Privacy and Federated Learning are ML and AI-related matters that unfold without exposing personal data. With the blessing of differential privacy and joint learning technology, Google is able to carry out artificial intelligence (AI) and machine learning (ML) without exposing user privacy.
Fitzpatrick also talked about privacy issues, especially advertising data, "We never sell your personal information to anyone. "No Gmail advertising, no sensitive information advertising.
Jen Fitzpatrick appears to introduce it further. Fitzpatrick said: "Everything we build is safe by default. We build everything with default security, including AI-driven security technology, photo encryption, Google Play Protect, and more."
"We believe that the first step to protecting user privacy is to have relatively advanced security guarantees." Obviously, this is a new attempt that cannot be ignored, that is, to take the hands away from the collection of user data. Fitzpatrick says the biggest security issue is still the wrong password use. Google-driven or phone-based authentication will help enable a "password-free future."
Pichai continues google's security progress in blocking malware, spam, and cyberattacks. There is also automatic deletion – deleting old user data. Pichai said Google has "easy to use" privacy controls.
Michael Peña was invited to visit Google's Quantum Data Center. Although the research is still in the early stages, Google clearly hopes to make its dream come true one day in the future. Based on current material and experimental site limitations, quantum computing must operate at extremely low superconducting temperatures.
Next, we will enter the Quantum Computing section. Even cosmic rays from outer space can disrupt quantum processes. Google is trying to reduce the error rate of its physical qubits, and it wants to build a quantum computer that can correct errors. Google has a new data center and quantum AI campus dedicated to achieving this quantum computing goal.
At the same time, Google's tensor computing unit (TPU) also developed to the fourth generation (V4). A unit can contain 4096 V4 chips, with a bandwidth of 10 times that of other network technologies and a floating-point hash rate of up to 10^18 (Exoflops). Pichai said they have already deployed and will have more carbon-free energy sources that can achieve 90% in the future. Apple has gained all the attention for its custom chips on consumer devices, but Google's cloud chips for machine learning are just as important, except that it abstracts from the hardware in everyone's hands.
For example, you can use it to search for something in a video, such as a momentary image of a roaring lion in the sunset. Multimodal technology is expected to be used in Google Maps or YouTube videos.
Pichai said: "Since it is still in the early stages of research, it is impossible for everything to be correct. "Pichai seems to like to show off the big research ideas that Google is doing, even if it's not a usable product. He did the same thing to Duplex. Pichai notes that now LaMDA only works for text, but he wants it to be "multi-modal" so it can understand images, audio, and "concepts."
Pichai explains how it works, having learned certain "concepts" from training data and then using "natural reactions" to keep the conversation going. After Pluto, Google also came up with the example of a smart conversation called Paper Airplane. The examples offer advice on how to fly farther, and there's a discussion of "what are the benefits of paper airplanes," and LaMDA is like being possessed by physicists and Aristotle.
Take BERT, for example, which basically adds contextual association to computer language recognition. LAMDA: A language model for conversational applications. A smarter response experience allows natural language conversations to continue more uncontroveringly. Google says it's open to any theme in all areas, not just simple things like restaurant reservations.
While Google's official rhetoric seems "mediocre," the search giant is showing off how they do things that are technically difficult to do, such as language translation, photo gallery classification, and video subtitles. Pichai also talked about a range of speech and image recognition technologies. He then turned to natural language understanding, which included converter models and BERT. "We have revolutionized the ability of computers to understand natural language," Pichai said.
Pichai returned to the stage and began to talk about search, especially the understanding of language and context. The Google Assistant has a commentary mode, and it is getting closer and closer to the "universal translator" in the user's pocket. Next, there are improvements in image recognition, Google Photos, etc., and it is expected that there will be more actions in the future.
Companion mode: Each remote user will have their own "video sticker," which will be released later this year. It looks like you can also assign tasks to someone directly in Google Doc via Google Task. There will be a "Custom View" in the Meet. Users can rearrange the position of people in the grid in the video, and also support live translation. Soltero says these features are open to all users, not just enterprise Workspace users
Soltero shows off a new feature: templates are auto-populated. When a checklist is in the process of being created, assistive writing suggests changing Chainman to Chainrerson. Starting today, you can easily showcase the files, sheets, or slides you're working on on the Google Meet cloud video platform. Later this fall, you'll even be able to implement Google Meet calls within Google Doc. Obviously, in the context of telecommuting, Google is working hard to integrate google Meet buttons into more of its own products.
Javier Soltero introduces Smart Canvas. For more than a decade, Google has been driving the continued development of digital documents. One of the new features introduced today is support for autofill of templates.
Chromebooks already dominate the space of educational devices, and Google will be talking about the "future of work" today, including new ways of working from home and the collaboration tools needed to complete this transformation. Pichai also mentioned Google Workspace, the new name for G Suite. Google Workspace is the new name for the previous G Suite. Google will launch a new work experience today, which is called smart canvas.
"Building a more useful Google for everyone" will be the theme of today. For several years, Sundar has been using "helpful" as a framework to understand Google's consumer business. The human-machine friendly interaction of predictive computing is driven by impressive artificial intelligence and data mining technologies. In Google Maps, there are now 150,000 kilometers of bike paths. It will soon offer an eco-friendly route that reduces emissions and a safer routing feature.
Google now has built-in tools to help people around the world easily search for vaccination sites. Pichai first introduced Google's epidemiological support efforts, including how it provides "high-quality information" in search. Google now has built-in tools that allow users to find vaccine sites in maps and searches. Google Maps offers more environmentally friendly route navigation options.
Pichai pointed out that the experience of the epidemic is very different, with the United States getting better and worse, and Brazil and India being much worse. Google will continue to maintain social distancing, masks and other protective measures. Sundar also spoke of COVID-19 remaining in an extremely destructive state in his home country, India.
Google has been preparing for the keynote for two years, and today's launch is expected to last 2 hours (followed by a developer session). The camera cuts to the Google campus in Mountain View, and CEO Pichai appears. In strict compliance with the COVID-19 policy, Sundar Pichai came to the center of the circular stage outdoors. Only a small number of people were involved at the site and a safe distance was maintained.
Some backstory about blobs: These blobs sing their "thought" opera sounds based on what the algorithm learns in training. Another model is responsible for achieving coordination. Created by David Li and Google's arts and culture team, Blob Opera isn't just a cute toy — it's a great example of how machine learning can be used to create something new and unexpected from existing data.
Judging by Pichai's tweets, the stage will be set in Google's campus, not the Shoreline Amphitheater. Now, the press conference performance has begun. Google played a real artist performance before keynote, rather than the traditional practice of playing music and ads.
In the description, it reads: "You're watching a machine learning music experiment consisting of Tune-Yards and Blob Opera." Commenting on the foreign media Outlet The Verge, "Google was very clever to put on a wonderful, real artist show before its keynote, rather than just playing music and ads, which prevented me from disgusting and satirizing its music choices or ads. Well, then humans are communicating with artificial intelligence. They are interacting with each other in some way. It's cute, but also a gimmick. ”
The hot field activity has begun
This year's I/O Developers Conference was published online, with animations and background music looping on YouTube's official live channel, all played by piano and marimba.
● More about Android 12
Usually, google announces many features and features of the new version of Android at the annual I/O developer conference, and we expect that this year will be no exception. Google has released some developer previews for Android 12, most of which bring a lot of detailed tweaks and developer-centric features. But at I/O, Google is likely to disclose some of the user-facing features in Android 12, including some rumored major user interface changes, such as stacked gadgets and a new lock screen with larger clock text.
Leak god JonProsser released screenshots of some Android 12 OS updates and said the new version of the system is expected to offer a new "beautiful new experience." This includes a series of new gadgets, buttons, and new animations, some of which we can see below.
With a new music widget, a redesigned Wi-Fi and Bluetooth switching interface, a new volume slider, redesigned notifications, and a new clock and weather widget that can be grouped and even implemented between different apps, Google aims to make the design of the entire system more cohesive.
● Upgrades to Google's main services
At previous I/O developer conferences, Google also liked to share major updates to its services and software. Over the past few years, we've seen Google use I/O to announce things like Google Maps' stealth mode, the new Assistant Voice, and the Google Duplex demo, in which Google Assistant calls a hairdresser to make an appointment. Google may be planning to roll out some similar exciting new features.
● Wear OS update
At present, Wear OS has come a long and bumpy journey. While Google continues to enrich its capabilities, the platform doesn't have a deeper thinking about the smartwatch industry. Over the past year, Google has added a new keyboard that supports third-party tiles and improves performance. The last major design change for Wear OS was in 2018.
Google's acquisition of Fossil's smartwatch technology in 2019 and plans to acquire Fitbit in 2020 have raised hope that it will also gain renewed attention to the concept of wearables. In particular, the process of absorbing Fitbit was stretched out because of the health data privacy concerns that accompanied it, but the deal was finally approved in 2021.
● Pixel Buds A series
We already know that the company is working on a new Pixel Buds true wireless headset, and maybe they'll be unveiled at this year's I/O conference. Google confirmed that the headset will support Google's quick pairing experience and deliver "high-quality sound." As part of the A-Series, the headset is expected to be cheaper than Google's current $179 Pixel Buds, but still comes with touch controls.
● Pixel 5a
Google officially announced the Pixel 5A 5G in April in response to rumors that the phone was canceled. But we probably won't see more information at this year's Google I/O conference. When Google announced the upcoming phone, the company said it would be available in the U.S. and Japan later this year, in line with last year's A-Series phone launches. The Pixel 4A, 4A 5G, and Pixel 5 were all announced in early August last year, so it looks like we may have to wait until sometime before we see the official exposure of Google's next mid-range phone.
● Custom processor
There are rumors that Google will develop its own chip, which will be used in its own Pixel model. Developers have found traces of Google's self-developed chips in Android open source. The developers found two keywords in the Android source code, Whitechapel and P21. Among them, Whitechapel is Google's self-developed chip code name, and P21 may be Pixel 2021, which refers to the model that Google will release this year.
According to pixel naming conventions, the model to be released by Google this year is the Pixel 6, so the P21 may be the Pixel 6, so it seems that the Whitechapel chip will be used on the Pixel 6. Previously, Some Google employees hinted that Google's new generation of Pixel mobile phones will use Whitechapel chips, but the specific details of Whitechapel chips are not known for the time being.
● There may be some surprises
Although Google's level of secrecy about new products is not very strict, Google will always bring us some surprise new products at I/O conferences.