Google Artificial Intelligence Assistant (Google's merger of two AI departments... AI battle in

 

1. Artificial intelligence software chat GPT

Reporter: Wen Qiao, Li Menglin, Cai Ding, Tan Yuhan, Editor: Gao Han, Lu Xiangyong. The competition for generative artificial intelligence (AI) sparked by ChatGPT is raging, and Google is struggling to catch up with local time. On April 20th, according to Reuters, Google's parent company Alphabet Inc. announced a major restructuring that will merge its two main AI research departments - DeepMind and Google Brain at Google Research Institute, To help the company gain an advantage in the AI field of competition.

2. Artificial Intelligence Movies

Google's impatience has become increasingly evident, while on the other hand, Microsoft's layout on large language models has become increasingly clear with the constant exposure of media - as early as around 2019, Microsoft began to contemplate plans to replace NVIDIA chips. In the face of chip shortages and high costs, Microsoft's choice of self-developed chips is indeed a way out.

3. List of artificial intelligence stock leaders

In fact, Google also released its first generation of TPU chips as early as 2016. An unnamed Silicon Valley engineer told the Daily Economic News that Google's models are all trained using TPUs, and compared to Nvidia's GPUs, the advantage lies in the particularly high chip level interconnectivity. Given the current high loss of power connected chips, a potential solution is optical interconnection.

4. Artificial intelligence AI

The AI war is becoming increasingly fierce, and both Google and Microsoft are using all their means to stay ahead in this competition. However, recently, the words of OpenAI CEO Sam Altman have caused an uproar in the industry. At a recent event at Massachusetts Institute of Technology, OpenAI CEO Sam Altman stated that the era of giant AI models is coming to an end.

5. Artificial intelligence computing power

In fact, what Altman said is not unreasonable. In order to help OpenAI train ChatGPT, Microsoft spent hundreds of millions of dollars tailoring supercomputers for it. Such high costs have already meant that this is only a game for a few people. The Silicon Valley engineer told reporters that the industry should not only focus on current commercial interests, but also try to persuade several domestic giants to give up the idea of training big models to solve specific scenarios.

6. Artificial Intelligence English

I hope there will be less competition in this industry and everyone will go straight to General Artificial Intelligence. If this is a market consensus, there won't be 300 companies, but only a few companies with long-term prospects working in this direction, "he said. At the same time, as the AI war becomes more intense, even pharmaceutical companies can't sit still. On April 20th local time, American biotechnology company Modner announced on its official website that it has reached an agreement with IBM, We will collaborate to explore the use of next-generation technologies such as quantum computing and AI to accelerate the advancement of mRNA research.

7. Artificial Intelligence GPT

In this context, where is the way out for AI competitions? The Daily Economic News reporter contacted industry insiders such as Silicon Valley engineers to answer Google's major restructuring, introducing generative AI into advertising DeepMind and merging Google Brain from Google Research Institute. This is Google's latest move to catch up with OpenAI and Microsoft.

8. Artificial Intelligence Stocks

DeepMind was founded in 2010 and became famous for its AlphaGo artificial intelligence Go program, which defeated South Korean Go player Lee See. In 2014, Google acquired DeepMind for $500 million, and DeepMind has since operated as an independent department.

9. Artificial Intelligence Big Data

In 2015, Google co founder Larry Page announced the establishment of Alphabet with the aim of making new businesses independent of Google's operations. Alphabet referred to these new businesses as "other bets" and they still belong to Google's DeepMind, which has always been one of Alphabet's "other bets".

10. Ranking of the Best Schools in Artificial Intelligence

In the past decade, DeepMind's collective achievements in AI include AlphaGo, word2vec, WaveNet, AlphaFold, deep reinforcement learning, and distributed systems and software framework used to express, train, and deploy large-scale machine learning models, such as TensorFlow and JAX; And Google Brain has also delivered many highly anticipated projects, including Transformer, which is also the cornerstone of ChatGPT creation.

According to reports, Jeff Dean, former director of Google Research, who oversees the Google Brain team, will be transferred to the newly established position of Chief Scientist and lead the company's "most critical and strategic technology projects related to AI, Including a series of new powerful AI projects, Google CEO Sundar Pichai stated in a blog post on the same day: "With the support of Google's computing resources, integrating all these talents into a dedicated team will greatly accelerate our progress in AI.

”In addition, according to a report by the Financial Times on April 20th, Google also plans to introduce generative AI technology into its advertising business in the coming months, so that it can produce more complex advertising content, even comparable to professional content produced by advertising companies. It is reported that Google has already used AI technology to generate simple prompts in its advertising business, encouraging users to purchase its products.

Based on the content presented in the internal manuscript, advertisers can provide content related to specific advertising activities, such as images, videos, and text. Then, artificial intelligence will recognize and organize these materials, and generate an advertisement based on the target audience and the sales target required by the advertiser. Microsoft's self-developed chip secretly developed "Athena" for 5 years?.

Since the birth of ChatGPT, due to its alliance with OpenAI, Microsoft's AI business has been taking advantage of ChatGPT's momentum. As ChatGPT continues to thrive, Google has frequently demonstrated its "killer weapon", from chat robot Bard to integrating AI into the office system Workspace.

Google's impatience has become increasingly apparent since the "red alert" was sounded, while on the other hand, Microsoft's layout on large language models has become clearer with the constant exposure of the media - as early as around 2019, Microsoft began to contemplate plans to replace NVIDIA chips on April 20th local time. According to the latest report from The Information, it was actually about 5 years ago, Microsoft began secretly developing a chip with the internal code name "Athena", with a total of 300 researchers.

The reason why Microsoft chose to develop its own chips is simple, as computing power is too expensive. According to The Information, in order to help OpenAI train ChatGPT, Microsoft connected more than 30000 NVIDIA A100 chips and spent hundreds of millions of dollars customizing supercomputers for OpenAI.

At present, Nvidia GPU A100 and H100 are still the most mainstream GPU chips for training large models. According to IDC estimates, GPU chips account for 90% of AI's computing power, while Nvidia accounts for 80% of the GPU market. TrendForce, a market research firm, calculates that processing ChatGPT training data requires 20000 GPU chips. As OpenAI further expands the commercial application of ChatGPT and other GPT models, its GPU demand will exceed 30000 (the report calculates that A100 is the main one).

The strong market demand and lack of supply have led to a shortage of chips in the entire technology industry. "It's not that (A100 chips) are completely irreplaceable, but they have strong versatility and are very convenient and convenient for everyone to use, making them the most suitable for training," Qian Yu, a senior analyst at Jiwei Consulting, who has long been paying attention to the chip industry, told Daily Economic News.

Now, if Microsoft wants to continue developing commercial applications for GPT-4 and subsequent models, it will definitely require a lot of computing power support. According to reports, the cost of developing "Athena" is approximately $100 million per year, according to Dylan Patel, Chief Analyst of SemiAnalysis. The operating cost of ChatGPT is approximately $700000 per day, and the cost per query is approximately 0.36 cents.

If OpenAI uses Microsoft's self-developed chips, the cost can be reduced by one-third. It is predicted that Microsoft may apply the "Athena" chip on a large scale as early as next year. Can Indian CEO Nadella lead Microsoft back to its peak? When it comes to Microsoft's efforts in AI, it is inevitable to mention the third person at the helm of Microsoft - Satya Nadra, a 55 year old Indian CEO.

Since taking office as CEO on February 4, 2014, the market value of Microsoft under Nadella's rule has risen from below $300 billion to over $2 trillion. It should be noted that Microsoft was once the world's largest company by market value, setting a phased market value record of $620.58 billion in 1999. However, due to missing out on the mobile internet era, Microsoft's market value has gradually shrunk to below $300 billion.

It was thanks to the scalpel like internal and external reforms after Nadra took office that Microsoft's market value returned to the first camp and became CEO in 2014. Nadra turned to Microsoft's ship. He believed that "Microsoft should turn to 'cloud first, mobile first', no longer taking personal computers first, or even mobile phones first.

How to make Microsoft succeed in a 'cloud first, mobile first' world is a common challenge we face. "He cut off the unprofitable mobile business group and laid out the direction of Microsoft's 2B business, making Office, Microsoft Cloud, and AI the focus of company research and development. From then on, Nadella gradually led Microsoft to regain its industry position.

In just three and a half years, he helped Microsoft's market value increase by over $250 billion thanks to impressive company performance. In October 2021, Microsoft squeezed Apple out of the highest market value position for the first time in eight years, and it was in the middle of that year that Microsoft's board of directors unanimously approved Nadra as the new chairman of the company, one of the most representative Indian executives in Silicon Valley, Bill Gates, the founder of Microsoft, once said of Nadella: "Nadella is a recognized leader, who has strong engineering skills, business vision and the ability to unite people.".

In one word, the CEO of a technology company should have three characteristics: excellent technology, business acumen, interpersonal control ability, which can be said to be both "soft" and "hard". In fact, Nadra's growth path was not smooth. Born in 1967 in Hyderabad, India's fourth largest city, Nadra, her father is a civil servant and her mother is a university teacher.

At the age of 15, Nadra went to Hyderabad Public School, and then failed in the entrance examination of Indian Institutes of Technology (IITs). IITs was the highest dream of all Indian students at that time, ranking second. Nadra studied electrical engineering at Manipal Institute of Technology, and became the first generation of immigrants to the United States when he grew up.

During his time in the United States, Nadella chose to pursue a graduate degree in computer science. However, he did not attend Stanford University or Massachusetts Institute of Technology. Instead, he went to the University of Wisconsin, Milwaukee, to work at a technology company with numerous experts. His educational background was not particularly outstanding. After graduation, Nadella worked for several years at Sun Microsystems, a company that invented the Java language, and then joined Microsoft in 1992. At the time, Nadella was only 25 years old.

Although his educational background is not particularly prominent, Nadella has taken the path of differentiated competition. He has long expressed his unwillingness to become a pure technician, and Nadella's classmates once asked him, why not pursue a doctoral degree? He replied, "I want to be a PhD recruiter." In 1997, five years after joining Microsoft, Nadella completed his MBA program at the University of Chicago, taking a big step towards his ideal path.

Nadella later became the head of Microsoft's Enterprise and Cloud Computing department, serving as Vice President in Microsoft's online research and development department and Microsoft's business department. His leading Microsoft Azure enterprise business was highly successful, with Azure being the core of Microsoft's cloud computing department, second only to Amazon AWS in the cloud infrastructure market and leading Google.

Azure has become a new revenue generating tool for Microsoft, maintaining a revenue growth rate of over 90% every quarter. Under this wave of AI led by ChatGPT, Microsoft was able to take advantage of it first because Nadella took over the olive branch that OpenAI had previously thrown, and since then, Microsoft has invested approximately $13 billion in OpenAI.

At Microsoft's annual shareholders' meeting in December last year, Nadella also stated: "Even ChatGPT, one of the most popular artificial intelligence applications today, is trained on Azure supercomputers." Will the giant model end optical interconnect chips as the way out? For technology manufacturers eager to get a piece of the pie, the problem now is that they cannot buy chips.

For mainstream chips currently training large models, such as NVIDIA GPU A100 and (newer generation) GPT H100, how many can a company grab in the market The aforementioned engineer stated that in the face of chip shortages and high costs, Microsoft's choice to develop its own chips is indeed a way out.

In fact, Google released its first generation of TPU chips as early as 2016, which significantly reduced power consumption and cost compared to GPU solutions. The aforementioned engineers told reporters that Google's models were all trained using TPUs, which compared to NVIDIA's GPUs, The advantage lies in the particularly high level of interconnectivity at the chip level. A TPU's Porte is equivalent to a cluster with a high degree of internal interconnectivity, which has thousands of chips and can provide very fast computing power.

Nvidia's GPU is currently not feasible, but they are actually developing in this direction, "he said. In addition, considering the high loss of current power connection chips, a potential solution is optical interconnection." That is, using fiber optic connections, the loss is basically acceptable, and the delay is also very low. "The aforementioned engineer said that for domestic chip companies, this is also a direction to consider for development.

It is important to focus on the current or future when making technology choices. Domestic chip companies can learn from Google's idea of computing power requirements for the next 5-10 years, refer to Google's TPU design, and consider the path of optical interconnection. "He told reporters that as the AI war intensifies, domestic technology manufacturers are using all their means to stay ahead in this competition.

But at a recent event at MIT, OpenAI CEO Sam Altman stated that the era of giant AI models is coming to an end. He stated that further progress in AI technology will not come from making models bigger. "I think we are at the end of the era of giant models, and ultimately we will make them better in other ways.

”This statement has caused an uproar in the industry. After all, since the birth of ChatGPT, there has been a frenzy of generative AI and large-scale language models both overseas and domestically. Abroad, Microsoft and Google are in a fierce battle, and many financially strong startups, including An topic, AI21, Cohere, and Character. AI, are also investing huge resources to catch up with OpenAI.

In China, since Baidu released the "ERNIE Bot" in March, many Internet giants, AI enterprises and start-ups have announced their own big models or computing platforms, and many people have sighed that China's Internet industry has not been so "involved" for ten years, but in fact, the high cost of training models has already doomed this to be a game for only a few people.

The aforementioned Silicon Valley engineer also told reporters that the industry should not only focus on current commercial interests, but he is also trying to persuade several domestic giants to abandon the current idea of training large models to solve specific scenarios and instead head straight towards the direction of GPT-5, Striving towards the direction of universal artificial intelligence does not mean that I hope all domestic companies can train universal artificial intelligence. What I hope is that there will be less competition in this industry and everyone will go straight towards universal artificial intelligence.

If this is a market consensus, it may be just a few companies with long-term vision working in this direction, rather than hundreds of companies making big models, "he explained演愈烈，就连医药公司也坐不住了。

当地时间4月20日，美国生物技术公司莫德纳在官网宣布与IBM公司达成一项协议，将合作探索使用量子计算和AI等下一代技术，加速推进mRNA的研究据悉，IBM的人工智能模型“MoLFormer”可以帮助科学家们了解潜在mRNA药物的特征，两家公司将结合最先进的配方与生成式AI来设计具安全性和有效性的mRNA药物。

根据两家公司的协议，莫德纳将加入IBM量子加速器计划和IBM量子网络IBM方面则将向莫德纳提供量子计算系统的访问权限，协助其探索和创造新的mRNA疫苗和疗法新冠疫情暴发后，莫德纳因其研发生产的mRNA疫苗与辉瑞疫苗成为全球最主流的一批新冠疫苗而名声大噪。

随着全球对新冠疫苗的需求放缓，莫德纳正试图扩展mRNA技术平台的临床应用以治疗其他疾病据报道，近年来，随着量子计算和AI技术与生物医药技术的结合不断深入，极大地促进了生物基础科研的进步和药物研发的效率去年，谷歌AlphaFold人工智能软件成功预测了人体几乎所有的蛋白质结构，意味着人工智能开始攻克生物科学和医学领域的重大难题。

每日经济新闻