Deepmind’s losses continue. Recently, the world’s top star company submitted its latest financial report to the British companies registry, which showed that deepmind had a loss of 477 million pounds (about 4.2 billion yuan) in 2019. Compared with the loss of gbp470 million in 2018, it increased by 1.5%. < / P > < p > deepmind is an artificial intelligence start-up company, which has developed many star products in the past 10 years. For example, alpafold, which was just launched recently, uses AI to predict the three-dimensional structure of proteins, and overcomes 50 years of challenges in biological science; alphago, an AI go player launched in 2014, defeated Li Shishi, a world go champion and professional nine stage player, in 2016, and Dang in 2017 Today’s first go player Ke Jie. < / P > < p > but behind these halo products, deepmind has never explored a feasible commercialization path, with a large amount of R & D investment for a long time, making it in a state of continuous loss. < / P > < p > Humayun sheikh, the co-founder of deepmind, once said, “if it hadn’t been for Google’s $600 million acquisition, deepmind’s AI Lab might have gone bankrupt.” < p > < p > since 2014, Google has paid for all of its R & D losses. However, under the long-term losses, this year’s deepmind also seems to have some good financial news. < / P > < p > according to the latest financial report, although its losses continue to expand, compared with previous years, the growth rate of losses has decreased, and its revenue has increased significantly – the revenue in 2019 is 266 million pounds, which is double compared with 103 million pounds in 2018. < / P > < p > let’s look at this year’s data again. In terms of revenue, deepmind’s revenue in 2019 reached 266 million pounds, an increase of 158% compared with 103 million pounds in 2018, the largest increase in recent years. < / P > < p > in addition, in terms of losses, deepmind increased by only 15% from 470 million pounds in 2018 to 477 million pounds in 2019, which was significantly smaller than 221% and 56% in previous years. < / P > < p > overall, compared with previous years, deepmind’s revenue growth is obviously accelerating, and the loss is also tending to be stable. It seems that deepmind is expected to turn losses into profits in recent years. First of all, the financial report mentions an important item: the turnover and R & D remuneration of other group enterprises. The data shows that most of deepmind’s main customers come from other companies under the parent company alphabet, mainly Google. At the same time, deepmind’s R & D achievements are mainly applied to these enterprises. For example, its AI is applied to voice assistant and data management center tasks by Google. < / P > < p > secondly, deepmind’s expenditure in 2019 reached gbp717 million, an increase of 26% compared with gbp568 million in 2018. According to the report, the increase in its costs “is mainly related to the increase in technology infrastructure, staff costs and other related expenses.”. This is very important. < / P > < p > although Google has not disclosed how much it charges, it is likely to rent deepmind at a discounted price, which means that if Google’s technology and service support, the cost of the company’s “technology infrastructure” will be much higher. In addition, the main research field of deepmind is deep reinforcement learning, which requires very expensive computing resources. It is understood that its 2019 development projects, including the AI system of StarCraft 2 and the AI system of Raytheon hammer 3, all cost millions of dollars in training costs. Alphafold, which was launched this year to predict protein folding structure, is also a very expensive project. < / P > < p > the employment cost of talents is also an important item that can not be ignored. In the past few years, although the number of talents involved in machine learning has increased significantly, there are very few top scientists who can engage in cutting-edge AI research. In order to compete for these top AI talents, large technology companies such as Google, Facebook, Amazon and Microsoft have launched a talent arms race. < / P > < p > it is understood that worldwide, the salary of top AI talents has reached 7 figures. Deepai has more than US $10 million in annual salary, and many of them are among the top scientists in the world. Most of these top people come from the world’s top universities such as Oxford University, Cambridge University, Stanford University or MIT. Some analysts believe that without Google’s support and investment, deepmind lab can’t hire more senior AI talents for their projects. < / P > < p > therefore, in general, although deepmind shows signs of a slow turnaround, its growth still depends on Google’s financial resources and large cloud infrastructure. < / P > < p > as mentioned above, Alpago, alpafold, etc. The former has achieved unparalleled accuracy in protein structure prediction, which is expected to solve the huge challenges of biology and computer science for many years. The latter defeated the world’s top human players many times in go. < / P > < p > in addition, in terms of multi-agent, the StarCraft AI alphastar developed by deepmind has also defeated the world’s top players for many times. At present, it has reached the master level in qualifying and has more than 99.8% of human players on European servers. < / P > < p > it should be noted that compared with go, real-time strategy (RTs) games such as StarCraft pose greater challenges to AI due to their more complex scenes, wider operating space and higher frequency of real-time decision-making. Therefore, compared with Alpago, alphastar shows stronger strength in simulating reality and playing games with human beings. < / P > < p > and since Google acquired it, it has been very satisfied with its technological achievements. For example, in a recent earnings conference call, alphabet CEO Sundar Pichai once said: < / P > < p > “I am very satisfied with our progress in AI research and development. It’s important that we’re a leading technology company and we’re in the lead right now. I’m proud of Google and deepmind. ” < / P > < p > in addition, Google also gave deepmind a written assurance that it “will continue to provide sufficient financial support to the AI company for at least 12 months.”. < p > < p > according to the financial report of ALP deep holdings, the company owes more than 1.1 billion pounds to its subsidiaries. < / P > < p > but in contrast, deepmind may be slightly different. Strictly speaking, deepmind is an artificial intelligence laboratory, which pays more attention to the research and development of cutting-edge technology than commercial exploration. < / P > < p > demis hassabis, CEO and co-founder of deepmind, boasts: “deepmind has established a world-class interdisciplinary team composed of machine learning experts, neuroscientists, engineers and ethicists, creating a unique environment. We will continue to invest in the research and development of cutting-edge technologies and look forward to bringing more breakthroughs to the scientific community in the future. On the one hand, scientific research is measured by decades. Nowadays, many AI technologies in commercial applications are developed from 1970s to 1980s. Similarly, many of the cutting-edge research and technology currently proposed may not be able to enter the mass market in the next decade or so. The ultimate goal of deepmind is to realize general artificial intelligence (AGI). The most optimistic estimate is that there are still decades to go. < / P > < p > on the other hand, investor patience is measured in years and months. If there is no profit or obvious growth hope in a few years, it is difficult for investors to invest in capital support. Obviously, the current deepmind has no obvious growth trend, its main customer is still Google, and its technology commercialization prospect is not clear. < / P > < p > this is the dilemma of deepfake. In essence, deepfake is a research organization, which hopes to break through the limits of science and ensure that AI is beneficial to all human beings. However, Google and its investors want to make products that can solve specific problems and make profits. < / P > < p > these two goals are diametrically opposed. Some people in the industry believe that although deepmind has made a number of major technological breakthroughs, there is no need to worry that its research results are unprofitable. Google also expresses its willingness to continue to pay for it. < / P > < p > but in the long run, its survival and development are more and more related to the interests of investors. Deepmind should think more deeply about its own future and the future of AI scientific research. Global Tech