Shulou(Shulou.com)11/24 Report--

Alphabet, Google's parent company, has closed up more than 4% in the past two trading days, its share price has quietly rebounded to its highest level since august last year, and its market capitalization has once again risen to $1.43 trillion. It seems that the AI report card handed over by Google at the I / O conference has been approved by the capital markets. And this is the message that Google executives hope to convey.

(Google closed up more than 4% for two days in a row) AI became the absolute protagonist I / O conference keynote speech began more than 80 minutes, the previous absolute protagonist Android system debut, and only in a hurry. This was unthinkable in previous conferences. In fact, the pace of Android updates has gradually deviated from the I / O conference. The developer version of Android 14 was released in February this year, and the first public beta version was launched last month.

Even Android's mouth-watering dessert code every year doesn't care much anymore. After all, healthy eating is the trend. By the way, the code for Android 14 this year is Upside Down Cake, a dessert that bakes stuffing at the bottom of the cake.

Looking back on the scenes of digital enthusiasts looking forward to the new version of Android every year, although the total number of Android devices now exceeds 3 billion, the platform itself has entered a bottleneck period of minor repairs. In addition to the human features brought by AI, it is difficult for the platform itself to have a major upgrade.

At this year's Google I / O conference, the schedule has been reduced from three days to one day, with only one protagonist from beginning to end: artificial intelligence. This is not the first time Google I / O has become an AI technology show. as early as 2017, Google CEO Pichai shouted the slogan "from mobile first to AI first", gradually landing the deep learning AI technology in Google's products.

It is no exaggeration to say that this year's I / O has completely turned into an exhibition of AI technology achievements, with Android only at the end of the list, while other products such as Chrome are ignored. Yesterday's keynote speech, Google CEO Pichai onstage less than a minute of greetings, began to go straight to the topic to introduce AI. Almost all the new products released yesterday are permeated with AI technology, especially generative AI.

According to US media statistics, Google presenters said AI more than 140 times in their keynote speeches yesterday. It was said 35 times when the PaLM 2 language model was released, 42 times when the Bard was released, and another keyword generation (Generative) was said 30 times.

During the two-hour keynote speech, Google released 15 AI products with features in various areas: med-PaLM 2, Vertex AI, sec-PaLM, Gemini, Project Tailwind, Codey, Chirp, Duet AI for Google Workspace and Duet AI, to name a few.

(the new PaLM2 model significantly improves the performance of Bard) what products did Google release yesterday? In the information-intensive two-hour conference, the core products are the upgraded version of the large language model PaLM2, and the upgraded version of Bard based on the training of the new model, as well as the new features of Google search, office building, cloud services and many other products under the blessing of AI.

PaLM2 is an updated version of the PaLM generation released last year and is the counterpart to the OpenAI GPT-4 model. But the performance of the second-generation language model has improved significantly in text reasoning, coding and translation, according to Google executives. PaLM2 is not a separate product, but a basic training model used to support many AI functions and products of Google. It has been used to train 25 products and functions, including Bard, an AI conversation tool.

Three months after the launch of ChatGPT, Google launched AI conversation tool Bard in February, but Google lost $100 billion in market value in two days because Bard gave the wrong answer during the product's feature presentation. This detail can also be seen that the market attaches great importance to the performance of Bard, believing that it is related to Google's future competitiveness.

After the public beta version of Bard was officially launched at the end of March, it was questioned by the market again because of the obvious gap with the performance of ChatGPT. Moreover, the public test of Bard is a full four months behind that of ChatGPT. After the launch of Bard, OpenAI released a new generation of model GPT-4, which put Google under great pressure.

The significance of PaLM2 is to improve the performance of Bard. According to Google executives, the new model not only significantly improves the accuracy and availability of Bard, but also improves previously weak coding query capabilities (which is where ChatGPT is clearly ahead of Bard). In addition, Google also announced the full launch of Bard, will launch a public beta version of Bard in 180 countries and regions, and will soon support 40 languages.

For ordinary consumers, what attracted the most attention yesterday was the improvement of Google's core search product under the blessing of Bard. In the face of users' questions, AI-blessed Google search will automatically integrate many search sources to provide users with simple and clear answers. At the same time, Google will provide sources of information, as well as traditional search results to help users make their own choices.

(AI-supported Google Office component) WorkSpace, a Google office component, has also acquired the SIdekick feature supported by AI. Users only need to give an outline framework, and Google can help write business emails, work documents, and generate tables. Google has also opened its Bard platform to enterprises, and thousands of companies have created dialogue services based on their own information based on Google's Bard.

The Android system has also been enhanced by the AI function. The Magic Compose function of Google SMS can generate reply messages for users according to the context of the conversation, and the generative wallpaper function can generate three-dimensional images of the pictures that users like as wallpaper.

It is also worth mentioning that in view of the proliferation of fake news pictures generated by AI on the Internet, Google has also launched the image traceability detection function "About this image" to help users find the source of the pictures and better judge whether the pictures are true or not. This feature will be launched this summer, which is the equivalent of adding a photo-like exif to online images.

(Google hardware design and pricing slot is full.) by contrast, many of the hardware products released yesterday, including the first folding machine Pixel Fold, the home tablet Pixel Tablet and the mid-range phone Pixel 7a, have received little recognition and attention. After all, Google's hardware is only used for software performance display, the design, workmanship and configuration of the hardware itself are difficult to surprise, and the pricing is full of slots.

The pace of commercialization lags behind despite saying AI more than 140 times in two hours, but throughout the event, Google mentioned only one competitor's name: OpenAI. All of their new product releases remind viewers of ChatGPT and Microsoft. This AI competition is not only related to Google's core business competitiveness, but also determines Google's position in the future industry.

Indeed, the past six months have been the most awkward phase in Google's 25-year history. Google made its first large-scale layoffs in its 25-year history, announcing more than 10, 000 layoffs, or 6%, earlier this year. After expanding aggressively over the past two years, with its workforce soaring by 40 per cent, Google has had to cut staff in the face of a sharp drop in the US economy last year and a slowdown in the outlook for advertising spending.

To make matters more embarrassing for Google, the Internet giant, which is known for its leading technological innovation industry, has to accept the harsh reality that they have lagged behind their competitors in the commercialization of generative AI and must speed up to catch up with the combination of OpenAI and Microsoft.

Perhaps Google did not expect that it would play the role of a pursuer in the AI field. After all, Google started laying out early on on this track, forming a Google Brain team in 2010 and launching the first generation of deep learning structure DistBelief the following year. In 2015, Google launched the second generation of deep learning framework platform TensorFlow, which has a clear first-mover advantage in the field of AI development.

In 2017, when Pichai shouted "AI first", Google launched Cloud TPU, the large computing infrastructure for AI, which completed a comprehensive strategic layout in terms of development framework, development community, large computing infrastructure and user data, relying on Google's billions of levels of user data to improve and improve deep learning.

Even in the field of generative AI, Google developed it early, unveiling LaMDA, a conversational application language model that has been developed for several years, at the 2021 I / O conference. Measured by the number of AI papers published and the number of papers cited, Google is still well ahead of its competitors, publishing nearly 500 research papers since 2019, with 22 of the most cited papers last year.

However, in the process of commercializing generative AI, the industry giant Google has been pre-empted by startup OpenAI. This includes not only the start-up advantage of startups, but also the misjudgment of Google executives. Over the past six months, Google has paid the price for its slowness in this area.

In November last year, startup OpenAI launched ChatGPT, an AI dialogue tool for public testing, which quickly became the focus of the global technology industry. Almost every day, the landing application of ChatGPT in many areas, such as media finance, collaborative work, education and learning, will become the focus of social networks and technology media.

Google executives are still not paying enough attention. In December, Google CEO Pichai and Jeff Dean, head of the AI project, said at an internal meeting that the technology was not yet mature and that problems with a hasty launch would affect Google's brand image and industry reputation because of its large number of users.

But as Microsoft connects ChatGPT to its own Bing search and introduces generative AI in office components such as Office and Teams, Google also feels unprecedented pressure to release Bard ahead of time. The information errors in the Bard press conference, and the public beta version is obviously not as good as the performance of ChatGPT, which makes outsiders even more worried about whether Google has lost to competitors on the generative AI track. The downturn in Google's share price over the past six months is a sign of investor concern.

The competition was just beginning after the I / O keynote speech, and Google shares rose 4% yesterday, their best performance on the first day of the I / O conference. Many industry analysts also gave positive comments on Google I / O, believing that Google is still in the leading position of AI competition and can bring positive returns to the company.

Morgan Stanley analyst Brian Nowak wrote in an analysis that the bank remains bullish on Google, believing that yesterday's I / O meeting will help address AI concerns that are hotly debated in the industry. After a disappointing February launch in Paris, Google remedied the I / O and more fully demonstrated its generative AI strategy, Jeffries analyst Brent Thill. Oppenheimer analyst Jason Helfsein also believes that yesterday's I / O keynote speech defused recent investor dissatisfaction with Google's lack of AI innovation.

Maybe that's what Google is trying to convey at the I / O conference: despite its slow initial commercialization, Google is still the strongest competitor on the AI track and has the inherent advantage of billions of users and huge amounts of data. Google has more than 90% of the market share in search, which is a huge gap that Microsoft Bing and OpenAI can't approach in a short time. As Evercore analyst Mark Mahaney wrote, "We don't think there is only one winner in AI. We just think that Google lost the generative AI competition is wrong, and Google's years of investment in AI, especially machine learning, cannot be ignored."

(testers believe that the Google model is inferior to GPT-4 in language comprehension.) however, Ethan Mollick, a professor at the Wharton School of Business at the University of Pennsylvania, believes that Google's model pays more attention to translation and mathematics, but still lags behind GPT-4 and Bing in language tests, after a horizontal comparative evaluation of PaLM2 and GPT-4.

In two weeks' time, Microsoft will hold a Build developer conference in its headquarters in Seattle, which will further showcase Microsoft's technology in the AI field, its cooperation with OpenAI, and the new features of Microsoft's core products supported by AI. At that time, the arms race in the field of AI will begin a new chapter.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.