A large number of artificial intelligent startups enter IPO? Do you see the huge trading market?

In the view of venture capitalists, the startup of the artificial intelligence is already time to enter the open market, and they have made more, larger, and earlier. For artificial intelligence startup, it is now a good time to raise funds. Although many of the financing records of the startups have risen in 2021, the more powerful and intelligent startups, which seems to have a broader customer demand, including more young and very long investors.

The frontier data from CB Insights describes in detail a global artificial intelligent financing market that is active. Of course, market financing may be slightly different in different regions. For these data, the AppHire Ventures Partner Jai Das and Glasswing Venture Partner Rudina Sseri conducted deep-depth research and analysis, further clarifying the flow of funds and development prospects in today’s artificial intelligent entrepreneurial market.

Record of transactions

The report of CB Insights shows that in the third quarter of 2021, investors have invested $ 17.9 billion to the world’s artificial intelligent startup. Since the first quarter of 2020, the financing amount of these enterprises has risen in quarterly. The artificial intelligent startup of the initiative is 6.3 billion US dollars, frequent transactions and recorded transactions have prompted these companies to usher in. Dramatic performance growth.

The total risk investment in the start-up industry has continued to rise in more large-scale huge round finance and billion or even more investment amounts, and the artificial intelligent entrepreneurial market is no exception. In the third quarter of 2021, a total of 45 super round financing was obtained in the third quarter of 2021. This is a huge financing of 45 or more in three consecutive quarters. In the first quarter of 2021 and the second quarter, 47 and 46 financing transactions were obtained separately.

The large trading amount shows the super round growth rate of artificial intelligent entrepreneurial companies, which is compared to how different years ago. According to CB Insights’s data and TechCrunch measurement, in 2020, an average of 19.75 huge transactions were obtained per quarter in the field of artificial intelligence. In 2019, this figure is only 13.5 per quarter, 12.25 in 2018, 2017 is only 6.

It is worth noting that accompanying time’s transition and growth of medium-term investment, the early stage of transactions is currently declining compared to the transaction share of the entire artificial intelligence startup. In 2020, early artificial intelligent investment accounted for 65% of total transactions, while medium-term transactions accounted for 20%. By 2021, this number is 61% and 22%, respectively; in 2020, there are 2,412 rounds of A1 round financing in the artificial intelligence field, including 2,073 rounds of financing in the third quarter. From this investment trend, it is foreseeable that the total investment volume of human intelligent industry in 2021 will continue to rise, compared to early transactions, the medium and long-term investment amount will have a new high.

In fact, in the financing of artificial intelligence, the oversized financing has fallen to 70% peak in the first quarter of 2021, fell to 57% in the third quarter. Early and mid-term artificial intelligent entrepreneurial companies are also expanding. The average transaction scale of artificial intelligence startup has increased from $ 18 million in 2020 to $ 3.3 million this year. Considering that the average data may have deviation due to some big transactions, we are more concerned with the median of the transaction size, which is also accelerating growth, which has increased from $ 5 million in 2020 to 700 to date. Wan Dollars, the increase is about 40%. More specifically, to the third quarter of 2021, the annuality of the angel wheel is a median of $ 6 million, above 2020 US $ 4 million.

These data seem to help us associate, a batch of startups are growing at high speed, busy solving a scene of another technical issue. Glasswing’s investors SSERI believes that although companies have become more extensive in recent years in recent years, artificial intelligence and machine learning products are also welcoming and challenge in various fields, which will further promote market demand. It also puts forward higher requirements for integrated operations from sales and marketing to cyber security.

Creating a wider range of customer bases and more advanced applications means that startup needs to invest a lot of money. Combined with data, we can draw the following conclusions:

  • The huge trading of artificially intelligent startups is constantly accelerating, which means that the initial company group is born more than ever. Most of these companies will need funds to drive their own exit, rather than seeking enterprises soft landing.
  • From the rising trading value, the transaction volume can be seen that the early financing market of artificial intelligence startups is strong, and the expansion of the medium-term transaction share should give post-investment to sufficient reserves. In a foreseeable future, this will also make the IPO of artificially intelligent startups continue to rise.

Of course, at the same time we also need to admit that all artificial intelligence startups we have focused can reach expectations with the above.

Artificial intelligence is everywhere

Although we are more interested in future technologies and cutting-edge artificial intelligence, many of these companies are actually handling more monotonous tasks every day. Of course, this will not cut the enthusiasm of the wind. Sapphire Ventures Investors DAS is very optimistic about these "bored artificial intelligence", which are used to solve many simple problems. As an example, the company is providing transcription and subtitle services that rely on artificial intelligence, and has been committed to "humanized" management, and maximize the accuracy of work.

Standing in sapphire’s pragmatic perspective, it does not oppose human cooperation with the machine, which is known through its attitude toward Verbit and some healthcare companies. DAS believes that "artificial intelligence will not replace human beings, but will make humans do better at work. Artificial intelligence is outbreak out in all fields, I think any development software will have built-in artificial intelligence."

It is because of artificial intelligence becomes universal in people’s lives, and it is easier to create an artificial intelligent startup or investment. "In the past few years, the market gauge has been significantly reduced," Glasswing’s investor Sseri said, "AI / ML is no longer simple in depth technology, it takes two or three years to realize business feasibility." She It also tries to explain this with the extensive model (Openai’s language generator GPT-3), template and powerful open source community, trying to explain this, in her opinion, "startup can use them to generate innovative solutions."

At the same time, SSERI added that accompanying artificial intelligence has increasingly significant added value, people’s demand is increasing, "company, especially enterprises, have been put in their data infrastructure and practice More and more funds, they can now get huge profits now. As the value of the bottom line tends to be clear, suppliers with cutting-edge, high-value innovation technology may require the improvement of the product service price. "

This is also reflected in the supply and demand resonance map proposed by the venture capital company PI Ventures, and the value of artificial intelligence has a significant change in the two axes in the figure: as the technology is simpler, the supply of artificial intelligence is increased; As demand increases, the previous potential demand is now "surface".

From the perspective of venture capital, the results are obvious: "More entrepreneurs can enter this area, prove the attraction of the industry, and get more funds from investors. But if you are the founder, don’t expect The interests of investors have been fair allocation, at least yet now, "SESESERI said.

Geographical dilemma

When we study the flow of artificial intelligent funds, the situation will become uncertain when it is specifically areaful. Taking the United States as an example, in the third quarter of 2021, the American artificially intelligent startup company received $ 10.4 billion in investment, a total of 324 transactions. In Asia, the total financing of these 321 similar transactions is only $ 4.8 billion. The Europe is less, and it only has a value of 1.6 billion US dollars from artificial intelligence starting funds from 142 transactions. There are also Latin America, with only 12 transactions, worth $ 500 million. In fact, Canada won 24 rounds of financing in the third quarter, achieving $ 400 million in capital flow, and its artificial intelligent risk investment amount, almost all venture capital acquired in Latin America.

In contrast, it is not difficult to conclusion. If the startup of the initial company wants to raise labor intelligence funds, it must be established in North America, Asia or Europe, and the other markets are not active. Of course, if an early trading leads to a small market that has been born in a small market, the current situation may change.

Many times, we will see similar intergenerational differences in various entrepreneurial markets. North America, Asia and Europe may be farther in certain areas such as financial technology, which promotes the efficient solution of startups to a series of issues.

In contrast, emerging markets in developing countries such as Latin America and Africa have emerged in huge financial science and technology investments in recent years, which may be because of their average economic levels, and there is less interference from financial ecosystems. If this status is true, we should see that the number of financing numbers in Latin America and Africa in the next few years will rise.

Double twelve phones worth buying? Smart people will choose these

First paragraph: Red rice K40

This mobile phone double eleven is very hot, there are many people who have not grabbed it, and they don’t miss it again. This phone is equipped with the Snapdragon 870 processor, with 120 Hz refresh rate, and AMOLED screen, which uses Corning’s five-generation gorilla glass screen. The camera has a front 20 million pixel + rear 48 megapixel +8 million long focal microspace lens + 5 million pixels. There are 4520 mAh batteries + 33W fast charge. Supports NFC functions and infrared features. Overall, it is still very good, it can be said that it is the cheapest mobile phone for 370 processors. At present, 6 + 128g reference price is 1799 yuan. It is estimated that double twelve will fall.

Second article: iqoo neo5

This phone is also a hot-selling item of double eleven. The pattern is equipped with the Snapdragon 870 processor, but it has a separate display chip. This is why so many people choose it. There is 120 Hz refresh rate and full-coverage cold cooling system, with AMOLD competitive screen. Front 16 megapixel + rear 48 million pixel main photo +13 million pixel wide-angle lens +2 million pixel black and white cameras. 4400 mAh battery + 66W fast charge in the battery, 30 minutes can be full, and the battery life is very good. Support NFC function and screen fingerprint unlock. At present, 8 + 128g price is 2438 yuan. Double 12 estimates are around 2200 yuan.

Section 3: Red rice 9a

This mobile phone sales is the third place in this year, 8 + 128G less than 900 yuan can be bought, there is a top 5 million AI beauty camera and 13 million HD main photos. The 3D surface fuselage adds a high level. There are also 5000 mAh batteries and infrared features. The appearance is also very nice. Very suitable for useless elders, give yourself a backup machine. And during the double twelve will fall.

Quaternary: Glory 50PRO

This phone is very popular, and the double 12 will definitely fall. If you like to buy Huawei, it is also recommended to buy glory 60, double twelve before it will be officially sold. Glory 50PRO can also be purchased, it has a 6.72-inch OLED superclapp, 120Hz screen refresh rate + 360 Hz screen harness rate. There is a front 32 megapixel high-definition lens + 12 million VLOG video lens, the back is 100 million pixel super clear lens + 8 million wide-angle lens + 2 million macro lens + 200 million deep lens, the camera ability is still very good. There are 4000 mAh batteries +1 million super fast charge in the battery, 25 minutes can be full. Supports NFC features and screen fingerprint unlocks. At present, the price is 8 + 128G in 3199 yuan, and the double 12 estimation will drop to around 3,000 yuan.

Article 5: OPPO Reno7 Pro

This mobile phone is a mobile phone that has recently been put on the shelves. All aspects are very good, and the double 12-one price reduction must have a lot of people to buy. It is equipped with the Tianzhu 1200-Max processor, AMOLED screen, and the camera has a top 32 million pixel lens + rear 50 million pixel main photo +8 million wide-angle lens + 2 million macro lenses. There are 4500 mAh batteries and 65W super fast charge. The appearance is really very nice, and the details of all aspects are very good. At present, 8 + 256G reference prices are 3639 yuan, and the double 12 estimates will drop to around 3,400 yuan.

Laser radar industry chain adjustment

1. Technical Principles

Laser radar is divided into 4 systems: laser emission, scanning system, laser reception, information processing.The transmitted laser pulse is transmitted by the drive laser, and the direction and line numbers are controlled by the high speed controller.Performance and cost is mainly optimized on the light source.

The scanning system implements a flat-screen scan with a stable rotational speed or MEMS.

Laser main photodetector receiving system, receiving back the reflected laser target object, the received signal is generated.

The information processing system, signal processing and conversion requires amplified. Get target block calculated by the information processing system in accordance with the surface, scanning the object modeled.

Performance Evaluation: line number, azimuth, scanning frames, the angular resolution, accuracy, detection range and the like. The core index is the detection range and detection angle.

Conventional mechanical technology is relatively mature path, the principle is simple, the motor rotation + mechanical structure to achieve a constant speed in the horizontal direction by 360 degree scan, the vertical orientation of the light scanning distributed. Advantages of mature technology, scanning speed, high resistance to light interference, cost-effective, can be 360-degree panoramic scans shots radars. Disadvantage is that the number of lines corresponding to a laser transceiver element, the number of lines increases the cost rising exponentially; tens source precise positioning requires a certain experience in the installation manual ability, senior engineer can do so expensive. The possibility of low volume production, automated production is difficult, requires senior and experienced engineers; the need to increase the number of lines stacked devices, long production cycle, velodyne delivery cycle more than 6 months; after being challenged by increased reliability components. On behalf of the enterprise is velodyne, Valeo; domestic game is Wo science and technology and Sagitar.

Flash pulse light source using a disposable market coverage, and map calculation using ToF, many mobile phone camera with low cost. Power can not do much, or injury to personnel, used for close-scene, AGV, robots use more. Simple principle advantage is low cost, disadvantage is the large crosstalk. Car for blind compensation is a solution, but to be placed in front of 80-100 meters is not a good path.

From the car overall cost, performance expectation is to see higher the MEMS, is the principle of mass production of the mechanical structure on the chip imputation by microelectronics processes, a small volume of the micro-galvanometer, aluminum plus a light emitting structure on the chip, and scanning angle to realize a large detection range. Change the scan mode, the transmitting end and the receiving end do transformation optimization chip architecture, the advantage of light weight low cost mass production, a small amount of technical path of the transceiver unit, and 3d achieved by scanning galvanometer of MEMS to reflect the laser beam, for laser and detector needs significantly reduced, and low cost. Key focus for scanning an object can achieve compensation algorithm, the detection range of up to 120-200 meters, a good product can sacrifice angle reaches 300-500 meters, such as Luminar products. Technical performance disadvantage is a gap, there is tradeoff between detection distance and angle. Mechanical scanning beam can be increased by superimposing the light beam transceiver implementation, the laser can be increased to increase the distance. In close-up with the blind and the forward angle of less demanding scenes still ok. Only l4 there are more than 360 requirements for mechanical scanning laser radar, autopilot many depot system is a multi-system integration, visual and millimeter wave has, before the laser radar to 200 meters, the detection angle of 60 degrees can be accepted. Lidar companies are upgrading, detection range and angle of the two core indicators

Only the OPA pure solid, the principle of controlling the light source to control the phase difference by interference velocity direction, laser light by changing the phase of the electrical signal, by constructively or destructively interfere to form a single angle controlled, electric signals only need to scan different angles, without any mechanical structure to achieve the effect of mechanical scanning. Landing hard. Omitted mechanical scanning structure, to achieve a similar mechanical pan-scan technology is expected to be the final route, but the technical difficulties, the process is important to the stability of the bottom layer. Waveguide control OPA will need to change the underlying structure of the semiconductor material to achieve a breakthrough. Currently advantage is cost-effective, low power, the future good scanning performance; however laser commissioning, operation a large amount of signal processing, high component costs, achieve more difficult. The final solution also needs 5–10 years, upstream suppliers need to mature technology.

Mainstream technology path: mechanical, solid state MEMS Flash and so on.

Special Technical Path: Dajiang and Xiaopeng, Dajiang laser light source 1 is reduced, cost reduction achieved through a high degree of special scanning manner; Ouster significantly improve reception sensitivity by replacing the light source and the self-developed mm chip, to improve the detection path .

Overall technical path is no convergence, the car costs and weigh on core indicators, pointing in the basic mechanical, MEMS and other three options and more.

2, a laser light source, the detector cost structure and so on

Currently various different routes, laser, the MEMS galvanometer, chips, circuit boards transmitting end, about 30%, the laser floating large, depending on whether the 905 or 150 source, additional mems galvanometer suteng so do a lot of work on the galvanometer, galvanometer radius to lift from 1-2mm 5-7mm, receiving a reflected light receiving surface is large. The transmitting end is the major cost components.

The receiving end, the receiving plate and the chip, 20% -30%, fluctuate, because of the different chips.

Housing structure 10% 10%, optical components such as lenses.

MEMS assembly and adjust labor costs and the like 10%, 50% mechanical labor costs.

3, most of the plants with a light source 905, Luminar and 150 of generic light, two light sources the advantages and disadvantages? Optimistic about what?

The core difference in detection range. 905 light source detection range of 100-150 meters is very good, only 200 meters further to increase the transmission power of the human eye impact, all the way to see the laser radar vehicle, increase the power of the human eye would endanger security. The light source 905 mainly made up of 30-50 meters blind, unmanned AGV car, on the car headlights and other applications to detect a curb. In the words of the forward speed of the Program, Teng, but less than 200 meters.

150 is a good source core strengths detection range, Luminar and up to 150 general-purpose light source can reach 300-500 meters, another eye-safe better. However, the price is high, to a few hundred dollars, the light source 905 is equivalent to 6-7 times. A light source chip 150 of the receiving end to change the development and optimization do, have deep technical accumulation.

4, long-term costs to see what level to large-scale car?

The price of the laser radar factory is very high, but the cost will have a significant decline in 10,000 levels, and the laser radar manufacturer’s own plan, Luminar announced the price of 1,000 dollars, customers are Volvo and Model, the detection distance of more than 300 meters , Use 105 nm light sources.

Sagitar, optimized the detection distance through the spray mirror, with a 905 light source to achieve a 200-meter detection distance, priced at $ 1998.

Innoviz’s customers are BMW, the progress is slow, PRO1 and 2 are 250 meters, the angle is different, the prices of 500 US dollars, $ 550 meters of products.

Laser radar manufacturers are between 1000-2000 dollars, which is a preliminary mass production price.

The car factory generally is expected to be within 1,000 dollars in the cost of laser radar. Take a new energy car as an example, 400-500 US dollars fixed price, the car factory’s own expectation will drop to 200-300 dollars. When millimeter wavelene reaches 200-300 dollars, the laser radar is reduced to this number and the car factory is expected.

If you have a batch of cars, you are expected to have 500-1000 US dollars, and some manufacturers have a 500-US dollars to grab the market.

(More investigation content can pay attention to the public account "Investment and Research Front" Messages, or join the community, experience more 0 expected poor information)

5. How much is the amount of laser radar planning in the factory?

There are many models, the amount is not so good, can be converted in the new energy industry, such as luminar, customer Volvo and Model recently depicted a few cars. Domestic Sagitar and Hechuan set point. Xiaopeng shipment has accumulated 100,000 units. P5 has started plus laser radar, and the P5 high-end models are selected from the laser radar. It is supplied by Dasheng, high with 30% -50%, and now the shipments are 8,000 per month, monthly laser radar Shipment of 2,000-3000 units. Xiaopeng G9 is a sagitar, plus P7, 1-200 months, so laser radar is a magnitude of about 5,000.

Other cars must be cargured, and the cycle is relatively long.

6. Which is the final plan?

Tesla’s Mask idea is relatively simple, I feel that people drive against the eyes, the camera is better than the human eye. Can’t say anyone who is wrong, Tesla is very high, and the high-end game can be run. Tesla cost management is strict, and it does not feel laser radar on technological maturity and cost recognition.

Since Xiaopeng added laser radar, it was rapid. Domestic vehicle factory is more active laser radar, Wei to ES8 crash event, industry insiders believe that laser radar can identify construction marks, perhaps the laser radar can avoid it.

7. Among the A-share suppliers, the value of optical parts, barriers are high, can it benefit?

High industrial value, optical components such as light source are currently related to market paths have not converge, and in the process of improving the core indicators, it is a key component of promoting performance in the system. For laser radar, it will be high in the next year. Moreover, the light source is not only used, mapping, unmanned cars, and robots may assist in the future, where the light source is the core device, the industry chain value is high.

Optical benefit laser radar development bonus, currently short-deficit and technical paths have not been converged, technical indicators have improved space, and third party teams have good living space. With some car companies, Haoyu provides solutions. When the amount is small, there is no energy research and development, and the supplier will be cut, and the scale supply has market bonus.

8. Can the laser radar can do?

The automatic driving upstream is positioning and perception, and the perception is the use of camera millimeter waves, and the midstream is decision-making and execution, and different manufacturers have different programs.

First, the L4 automatic driving company such as Modely, Baidu, etc., the execution algorithm is in his own hands. Some manufacturers bought radar and other components to do L4 systems, and the algorithm is in the manufacturer’s hand, and the laser radar is OK as long as it is good.

Second, Xiaopeng and other new forces, the laser radar also requires the algorithm to be open. Xiaopeng uses the laser radar of Dado this year, and the cloud map is particularly uneven, and many difficulties have been found in the back of the automatic driving architecture. The new forces increase the investment in talent, and in-depth understanding of the automatic driving architecture, which has the requirements for laser radar operation. Overall Softness is required, soft aspects require information to be collected and matched quickly.

The third is that the large-scale car factory, such as Beiqi with Huawei, maybe after all car companies are Huawei, like this, Huawei requires a lot of hardware and software.

9. How to see the whole machine?

Haoyu purchases the upstream components to integrate, its core in the optical parts.

10, Xiaopeng change supplier specific situation?

Triad is a laser source. There are two fast-rotating prisms that make the optical path to achieve petal scan. The center position is particularly sparse, information collection, and the imaging docking architecture bring direct trouble, it is difficult to compensate for algorithm. This program is dropped, and only one of the light source is left, and the special architecture leads to a difficult algorithm to make up for defects. Xiaopeng P7 supplier is Dacheng, G9 supplier has not announced ..

11, Turda cost?

Tita is also around 1,000 US dollars. Tita’s drilling solution is also a way to reflect more optical path scans. 8 laser transmission and reception can achieve mechanical 96-line number scan, theoretically cost, but the saves is not as good as MEMS and Dashi Program, estimated to have 1,000 dollars.

12, US stocks have a few company share prices?

The company’s valuation system of the secondary market looks at two indicators.

Luminar is a solid state, a quantity of productivity than a mechanical, so Velodyne is only 15-20 billion US dollars, and Luminar is $ 15 billion.

From a model, Model announced that the forwardlim of luminar is visible to the naked eye. INNOVIZ is set to BMW, the progress is not smooth, and the business is slow, that is, 11-20 valuation.

OUSTER uses special technical paths, the benefits are extra-cheap, but the power of several watts is not realized, and can only increase the photosensitivity. It seems that the market has questions about the route, there are many light pollution, how to eliminate the noise of crosstalk, so there is only 1.1 billion US dollars after the listing.

13. The car factory drops to 200-300 dollars is the value of a single laser radar. What is the amount of bicycle?

L4 companies need 360 degrees of scan, but many car companies are integrated architectures, there are many millimeters waves, ultrasound, camera, laser radar only 120 degrees or forward, and do 360 degrees, stitching three The roof can also achieve 360 ??degrees, and the difficulty of algorithm is said. Overall price 500-1000 can be accepted.

14. There are many domestic cars, and they choose domestic suppliers. Why is the supplier?

Luminar does not have a prototype, can’t find related products, Silicon Valley friends say that the Luminar prototype with a seal to take the seal, Luminar has a military technical background, which may have this reason not to enter China.

Mechanical Today’s shipments will exceed Velodyne, product performance is still better than foreign suppliers, and domestic supplier products are good enough.

In addition, the process of incorporating the system has a complex algorithm docking process, and domestic engineers are more convenient to live.

15. If you go to the semi-solid and solid state direction, the integratedization is getting higher and higher, the transferred optical components will disappear, and the amount of optical value will fall, how to think?

Cycle mismatch, new energy cars visible to the naked eye is in the first year. It is high in the first year. The laser radar is rapidly increased as the core architecture market, and the laser radar manufacturer is hard enough. The car factory is not particularly sensitive. I will install it first, and I have a difference in science.

Everyone thinks MEMS is the most feasible, because solving problems such as mass production, performance can also be, it is the best to eat bonus solutions, and the cycle matches can be easily transformed.

OPA pure solid state has a lot of company layout, millet in pure solids and mechanical voters, OPA underlayer, chip development, algorithm program, the whole vehicle must improve, to the MEMS window is long, There is enough company to grow up.

16. Recomminal related companies?

The receiver Japan is better, and the MEMS technology has a lot of differentiation. Domestic manufacturers, the amount of laser radar has not yet, the transmission and reception chip does not go to a certain amount, and the laser radar manufacturer will do it yourself.

17. What are the parts and the foreign parts you have to master by the laser radar yourself?

It is still not yet mature, so I have to rank it in person. After millimeter righte is ripe, the components are basically extracted. After ripe, there is no need to control the components in your own hands. It is tested that the process mechanical architecture, optical path, software algorithm, etc. In the early days, it is an iterative improvement process. It is necessary to do the transmission and receiving chip, kick off the distance from others, and have high R & D fees can also be understood.

18. Where is the industrial chain breaking opportunity?

Laser light sources can be large, and light source manufacturers such as resonant cavity at first level need to have a good light source to reduce costs and enhance emission performance. The transmitting end is the most important cost composition, and there is more opportunities for the light source.

19, chip, laser cost proportion, new companies in the industrial chain or mature companies have high probability?

The first-level market is much more, the experts are biased towards the level.

The new manufacturer has a small head, and the amount is small to match the car factory.

There are many companies that want to do light source this year. There are many industrial laser manufacturers. It has been killed into the Red Sea. This year, it is automatic driving, the core is that the laser radar has stimulated market demand. Many entrepreneurs have invested huge research and development costs, and the team’s background is very professional, and the entrepreneurial company has natural window-oriented advantages.

The industry will have a massive production advantage in the maturity stage. But now technical indicators iterate, the technology route has not been charged, and better new companies.

20. The convergence time of the technology route?

Unlike semiconductor bodies, there are so many companies at home and abroad, may not converge into one or two, only $ 100-20 billion valuation is available when Volvo and Model’s Luminar is available, and the core vehicle factory has an opportunity. It does not converge from the technical path for 3 years.

21, industry spread trends, how does the industry drops to 200 US prolles, such as how to reduce the transmit-end receiving end?

The outer casing structure is not high, the decline is small.

Handling the general-purpose chip on the motherboard will have a decline in price, which can drop 10%.

The mass production can reduce costs, this year, when the laser radar plant is not large enough to use MPW, the TSMC is still caused to cause warp, this batch is abolished, and there is no good rate after mass production line. This problem is.

The transmitting end is an important component that can be reduced. The 905 nm light source is faster, and 150 is slower.

22. The laser radar, Xiaopeng P5 and Tesla are not particularly strong. What is the reason?

It is an accident that can feel the difference. The reason why high redundant reaps are safe, because each component has limitations, Tesla’s automatic driving rain is not open, the camera can not rain in rain, the millimeter wavele does not recognize the road sign, purely locate the tunnel Can’t, the laser radar is not in the rain. High reduction is for security and underlying safety.ε ?? ES8 If there is a laser radar accident, you can’t know.

23, is the laser radar is a transition plan?

The low-speed market ratio such as robots is cost, and there is alternative. However, the new energy car L4L5 is a problem that the laser radar is a core component, pure visual high fog, rainwater, etc. is difficult to solve. An example of the camera defect is that if the camera is purely relying on the camera, it will instantly account for the visual field of the camera. The algorithm is unable to solve, the camera has limitations. Therefore, millimeter waves, laser radar, etc. are indispensable components.

GALAXY RTX 3060 Metal Master MINI Graphics Card Tu Reward: All metal shell, small clever

The demand for small size graphics has never existed. The RTX 3060 of this generation is a GPU that is now well-sized with only 170W power consumption with nice game performance and only 170W. Previously there were ITX graphics cards such as MSI, IGAME RTX 3060 Mini OC 12G L graphics card, now another graphics card GALAKA also released their own ITX Specification graphics card – GALAXY RTX 3060 Master Mini.

GALAXY RTX 3060 Metal Master Mini’s overall size of only 168 x 115 x 40 mm, compared with seven rainbow Igame mini series, positive close to normal adult palm size. Its radiator housing is used as a die cast aluminum alloy, equipped with a metal reinforcement plate, which is quite durable, and the silver appearance is also filled with a metal.

This card still has a certain degree of shoulder, but compared to the previous MSI RTX 3060 Aero ITX OC graphics card, the card is much smaller, the front pattern cooperates with the black circular single fan especially like a eye, as adopted Ying Weida RTX 3060 GPU’s graphics card, talking about the eyes can’t help but remember the eyes of Ying Weida’s eyes, this unintentional join is a bit means.

And the biggest wonderful of the GALAKIMAR’s RTX 3060 Master Mini is that it is supporting fan intelligent start-stop technology. When the GPU is idle, the fan will automatically stop, reduce noise and power consumption, this feature is previously tested MSI RTX 3060 Aero ITX OC and IGAME RTX 3060 Mini OC 12g L do not have it on both ITX graphics cards.

Finally, the GALAXY RTX 3060 Metal Master Mini is equipped with a single 8PIN external power supply interface, like the public version of the TGP power consumption of 170W, and an HDMI2.1 and three DP 1.4A interfaces are also configured. GALAXY RTX 3060 Metal Master Mini also supports GALAXY 30 Series New Times Market, and users can monitor and adjust the graphics.

The youngest motivation of the Congress of the Congress of the Dialogue: It is discovered by AI to find the physics law, and the Nature Subscription

Wisdom ()

Author |

Edit | Indo

Scientific exploration, perhaps the ai field, one of the most "star sea" imagination.

Mysterious life password, everywhere, the power of power, chaotic molecular sports … Scientists go to success, explore simple, elegant, harmonious mathematical equations to describe the general laws of the world.

From the viewing cloud distinctive objects to the numerical climate model, from Newton, Einstein, a relatively discovered law, and leaving a concentrate in the process of human development. The world is so wide, there are too many mystery, not being explored and defined by human beings.

When AI grows into a key cutter that digs massive data information, it has opened a new door to explore the scientific issue.

The world’s top AI institutions are put into action, DeepMind uses AI to crack protein folding problems, Huawei cloud training tray ancient drug molecular model to promote AI new drug research, NVIDIA announced to create "digital twin global" to study climate change …

These decoded unknown explorations are also the direction in which Sun Hao is studying.

Ministry of Renmin University of China, Minister of Artificial Intelligence, Implicate Associate Professor, Sumao

Sun Hao is 33 years old. At present, Sun Hao is currently the youngest long-term associate professor and motion of Renmin University of China.

In October This year, Sun Hao team’s paper "PHYSICS-INFORMED Learning Of Governing Equation" is published in the International Academic Journal Nature Communications.

After a few years, he returned to the motherland, he wanted to use AI to unveiled more potential laws, find more reliable theoretical support for physics, chemistry, biological, engineering applications.

Sun Hao team published on the International Academic Journal Nature Communications in October this year

First, from"Architect"arrive"Equation detective"

Sun Hao is not a "AI + Science Exploration" direction.

His undergraduate studied in the civil engineering major of Hehai University. After graduating, he took full scholarship to the University of Columbia to study the master’s degree and doctorate of Engineering Mechanics. Subsequently, he was doing a Ph.D., 29 years old, and under 30 years old. Elite "scientific list, 31 years old is elected" Top Ten Chinese Outstanding Youth ".

During the study period, Sun Hao is the kind of play, and the students who will not die, I really like to swim. When doing a PhD, Sun Hao insisted on swimming every afternoon, and often playing table tennis, soothing to relieve stress, restart the brain.

Before and after 2015, Sun Hao’s research background is mechanics, doing some applications such as intelligent buildings, smart urban intelligent infrastructure, such as sensor monitoring and warning buildings, bridges and other building structures. When making software analysis, he began to contact artificial intelligence and machine learning methods for resolving relevant data processing issues.

In this process, Sun Hao found that data did not imagine so much, information is not so sufficient, if only data is processed, the previous basic mechanics law is not enough to solve some dynamic problems of the building.

At first, it was interested in making it. Later, as the study gradually extends, Sun Hao realized that this is not only mechanically related issues, weather forecasts, other systems in physical worlds such as weather forecasts, there are many other systems in physical worlds such as cell movement. There are similar problems.

So, is it possible to make a set of AI theory, it is possible to find a specific scientific law from the data, form a new theory, and can also take into account the basic theory we have learned before?

With this as the starting point, Sun Hao’s research direction began to transition to "AI + Science Exploration".

He believes: "The future scientific exploration should not rely on people. Newton wrote the second law of Newton, Cople wrote the law of Copler, they did have any people, it is a genius, it is a giant."

The unknown problem in the physical world is so vast, so far, human beings only solve the hoss of the iceberg. If there is still only relying on people’s power, how many geniuses do you still have?

"So we are thinking, with this, it is better to let Ai play genius, and AI will help us find this law and put this formula." Sun Hao said.

AI explores nonlinear partial differential equations for describing the complex reaction diffusion process of biochemical systems

Second, return to the motherland, use artificial intelligence to explore the scientific law

Since 2017, Sun Hao successfully passed the University of Pisberg, the Northeastern University of Northeastern University, and the part-time researcher at the Massachusetts Institute of Technology, Sun Hao chose to return to the motherland in this summer.

"The domestic scientific research environment has become much better than before." Sun Hao mentioned the significance of domestic research and support in policies and funds, and in the long run, he has more confident in China’s scientific research.

After taught by Professor Wen Ji, Professor Wen Jirong, he immediately decided to join a young Ai Academy, just created two years – China Renmin University of China.

Gao Hui artificial intelligent building

The Gao Hui Artificial Intelligence College was established on April 22, 2019. It is the second-level college, the founder of the Renmin University of China, and the founder of the Gao Yu Capital and CEO Zhang Lei’s support construction. The Chinese Academy of Engineering Pan Yunhe is the director of the Academic Committee, Professor Tejong Executive Dean, the teacher sequence implements "long hire".

"I feel that the teacher is very pleasant leader. I think I should join this place." Sun Hao believes that an excellent leader has a guiding role in the growth of young people. Whether it is anchoring the direction of the AI ??change, it is still necessary to have a feeling.

Apart fromleadershipIn addition, Sun Hao also is optimistic about the other three advantages of the Gaozheng Artificial Intelligence College:Vision, system, culture.

In his opinion, the Gao Hui Artificial Intelligence Academy has the development of the AI ??development and took the lead in proposing a temperature AI. Ai is not only a tool, it should be deeply associated with people, combined with various disciplines such as law, communication, economics, social governance, to solve extensive issues in the real world.

Can a place to attract passion, a soft young scholar, a system is critical, and the stereotype will give innovation to form a barrier, and the Gao Hui Artificial Intelligence Academy encourages the teacher to freely explore the research direction and does not set the boundary of the roll box.

In the next two months, Sun Hao felt that the culture of the college is very inclusive. The college supports professors to choose their own research direction. They have encouraged attitudes in internal and external cooperation. The research funds are also more abundant. If you want to do a research on future development, the college will specially apply for a part of the funds to specifically support This direction.

Current Sun HaoArtificial Intelligence Mathematical Basis and Improving CrossContains interpretative deep learning, knowledge characterization and reasoning, physical driver depth learning, symbol strengthening learning and reasoning, data-driven complex power system modeling and identification, control equation finding, infrastructure health monitoring and intelligent management.

After joining the People’s Congress Human Human Corporation this year, Sun Hao quickly formed a "Physical Driven Artificial Intelligence" research team, and published a papers in the international academic Journal Nature Subscription in October.

Up to now, he presided over or jointly hosted more than 330 million US dollars such as research projects; 50,000 papers were published in various publications such as international first-class SCI journals and computer topics.

Third, cooperate with the Meteorological Bureau to crack the weather password

The study of Sun Hao team advancing physical systems, chemical systems, biological systems, etc., to explore the various potential laws of thousands of physical worlds, and describe it in the form of mathematics.

At present, Sun Hao, Wen Ji, Lu Zhiwu, three professors and high-time high-end high-resolution data from China Meteorological Bureau to explore specific laws and equations with high-level high-resolution data, and to predict weather changes more accurately.

The traditional meteorological prediction method is based on the numerical calculation. According to the actual situation of the atmosphere, the fluid mechanics and thermodynamic equations that describe the weather evolution process are solved under certain initial values ??and boundary values, thereby predicting atmospheric motion state and weather in the future.

This exists a problem, the imaginary numerical model is in error, and the basic weather law can be captured, but it may be underestimated to some small probability of extreme weather.

How does these errors cause? Imagine models may be unreasonable, or people have not considered in place.

In this regard, Sun Hao uses AI methods, explores the laws with the actual meteorological evolution, and attempts to describe a better equation to achieve ultra-short weather forecast, and improve long-term prediction and extreme weather forecastion. Precision.

The Sun Hao team has three goals with the scientific system obtained by AI:Interpretative, push capability, generalization abilityEven if there is no event, the system can make a stable prediction.

This is quite challenging direction. The traditional AI model is a black box, but Sun Hao wants to change this concept, I hope to do "AI + small data" can still get a very reliable AI model, and can extract some unknown rules from small data.

AI explores nonlinear normal differential equations for describing chaotic power systems

This is also a very exciting direction. In scientific exploration and engineering applications, true valuable is not the amount of data, but not the information is sufficient. There is a lot of massive data, but even if there is no massive data, it is still possible to explore a simple and general law.

Mathematics is the foundation of all science, equation is an abstract expression of various scientific theories, and it is often a beautiful and beautiful equation that describes some objective laws of physical worlds.

In addition to meteorological precision forecasting, Sun Hao continued to pay attention to the progress of protein structure, automatic driving, etc., and he also considers the combination of automatic driving control systems and the like to optimize It can be interpretative.

Four,"To make big results, there must be big mind and vision"

In the College of Artificial Intelligence, Renmin University, Sun Hao is teaching graduate students "artificial intelligence interdisciplinary application thinking and actual combat".

"This class is not good." Sun Hao said that this is a new course, compared to existing artificial intelligence theoretical courses or case classes, it pays more attention to the guidance and training of students’ interdisciplinary research thinking.

For example, throw a problem, use ai method to solve it, how to embed it in the knowledge you need, and play a real role in designing the AI ??algorithm? This is the analysis capability that students should master.

He will also train students to study the articles of the top party, not only to understand the frontier method of solving the problem, but also clear how the story is talking about, and why they can make this method to solve the problem.

Sun Hao did not want to give students a limited time. When he guids Ph.D., he usually understands the students’ own ideas and career planning. Specify a direction for students who have just enrolled, but will not limit what method is used, gives doctoral students freely explored space .

"We solve this problem itself is very interesting, it is unknown world." Sun Hao said, "You make a regular, or find a new knowledge, it will become very excited, this feeling is constantly stimulating In the process, it will keep interest in interest, so it is said that there is such a trait that scientific exploration is why so many people love scientific exploration. "

He hopes that students must first love their research, clear what they want, do not have to be tangled this year, a nature is not half a year, "To make big results, there must be big mind and vision."

Everyone has different research traits, so Sun Hao will teach in accordance with the materials. For example, a student’s logic thinking is strong, learning and solving the problem, Sun Hao guides him to do some challenging things; and if a student does not have this potential, Sun Hao will slow down, come step by step. He started research.

He is happy to share your experience to the students. The doctoral student he is driving is very strong. When the students feel that the pressure is big, Sun Hao will stop the work of the hand, go out to climb the mountain and relax, and come back to see this problem, often appear new Inspiration.

At the same time, Sun Hao’s requirements for students have very high requirements: "If this student is reading a doctor, he needs to fully explore his potential and increase his research capacity to the ultimate of herself."

Do it, it is not to press the student, but it is hoped that the students can use solid basic skills after training, so that his future has a more wide direction of research. "Ai + Science Exploration" is essentially a research foundation AI method, even if students want to try some different directions, they can also swim.

For students, Sun Hao also expressed encouragement and support. In his opinion, in "AI + Science Exploration" direction, molecular effects or incentive new drug development, fluid model or drive rocket aircraft design, all kinds of applications such as this can produce value by increasing efficiency.

Conclusion: I look forward to more interdisciplinary talents.AIExplore the scientific star sea

Interpretation is the development direction of the future AI, and the combination of AI and scientific exploration is exactly an important carrier.

The current "AI + Science Exploration" is also facing multiple challenges: this is a region that requires depth interdisciplinary cross-disciplines. It is necessary to combine very strong expertise; in addition, how to make the system have both powerful learning skills, but also The beauty, this is a complex balance problem that needs to explore the new AI algorithm.

"We have made some work. I believe that in the future, we should at least have a role of guidelines." Sun Hao took some examples, including symbol enhanced learning reasoning strategies and sparse regression.

However, talents in the field of "AI + Science Exploration" are still quite scarce, and students are mostly born for computer classes. Sun Hao also hopes to see more other directions are interested in studying in this field.

"If mathematics is still good, basically think that the threshold of AI itself is very low, I am not a computer profession." He has only two basic requirements for his graduate students. First, there is enthusiasm for Ai + science, second is enough. Diligent, brave overcoming difficulties.

"Sweat and harvest are a pair of friends, diligence and knowledge is a pair of couples." He looked forward to more young people to join this direction, willing to pay sweat, to overcome a series of research problems.

February 8 release, Samsung S22 appearance exposure, small screen mobile phone "ceiling"

In the past years, Qualcomm flagship chip is handed over to Samsung or Xiaomi, the next generation of Snapdragon 8 Gen1 chip is about to debut, so everyone is very curious, the starting model is Samsung S22 or Xiaomi 12. More unfortunately, these two mobile phones have no throne 8 GEN1, which is said to be the right to Motorola Edge X grabbed the right, Xiaomi 12 series followed, Samsung S22 is to postpone until the beginning of next year. At this time, the digital blogger exposure news, Samsung S22 is expected to be released on February 8 next year and plan to officially launch on February 18.

After listening to this news, many Samsung’s death powder began to move, and the next generation of Samsung S22 is expected. According to the supply chain, the Samsung Note series will stop, and the Samsung GALAXY ULTRA and FOLD folding screen products are folded. Considering that most users of the folding screen mobile phone can not afford, everyone will put more attention on the Galaxy product. In addition to the date exposure of the press, there is also a Samsung S22 series of tempered film and lens configuration together, which is a small screen mobile phone "ceiling".

As shown in the figure, the camera specifications of Samsung S22 and Samsung S22 + are completely consistent.It is a 50MP main lens.The blogger also exposed the specific parameters of Galaxy S22 and SS2 + camera, mainly photographing 1 / 1.57 inch CMOS, supplemented with 10 million long focus lens, up to 3 times optical zoom, super wide angle pixels reached 12 million, the front camera is 10 million Pixels, camera parameters are not gorgeous, even the latest GN2 lens is useless. Of course, Samsung next will focus on calculating photography, even if the sensor size is small, the photo quality is not poor.

For Samsung, the screen has always been a proud selling point. The Samsung S22 of this generation is no exception. The front is equipped with a 6.06-inch AMOLED screen, 120Hz high brush design, intermediate punch design can bring a higher screen Compared with the 3D ultrasonic unlock design, the screen experience of Samsung S22 is good. Samsung S22 + screen size is raised to 6.55 inches, strictly in the big screen mobile phone, and larger size means to accommodate larger batteries, so Samsung S22 + endurance will be more powerful. At present, the entry version of Samsung S22 is the small screen hand, positioning the idiogram iPhone12 mini.

In the current market, there are very few small screen mobile phones, like iphonese2, iphone12 mini and other sales are not optimistic, even exploding Apple will stop more MINI version next year, and the future small screen mobile phone will be less and less. In this way, Samsung S22 is likely to be replaced, with a perfect picture quality and a lightweight harvest. Before the Samsung S22 is officially released, the old Samsung S21 is also a new round of price cuts, and the cost performance is obvious. Samsung S21 is equipped with 4000mAh capacity battery, and the weight is controlled near 170g.The peripheral parameters are full, OIS anti-shake, the bottom shot and IP68 waterproof is not absent.

Seeing this, many people are hesitant between Samsung S22 and Samsung S21. From the perspective of performance, the newly equipped is 4nm Snapdragon 8 Gen1 chip. The running score data will exceed 1 million, the same is 8GB + 128GB storage The performance of Samsung S22 will be more powerful. Of course, the old Snapdragon 888 can also take 800,000 runs,Don’t say that running the "Glory of the King", even if you connect the heavy game "God 3".With smooth one UI system, Samsung S21 still gives a full frame rate gaming experience, Samsung S21 price diving 1500 yuan, the lowest to 3499 yuan, suitable for the pursuit of price-effective users.

Talking about the last

Considering that Samsung S22 is equipped with the Snapdragon 8 Gen1 chip, the material cost is expensive. The starting price is expected to exceed 5K, one pair is more suitable for the "three-character" Samsung S21 is more suitable for the public, and the money is saved. A red rice K40 does it not incense? What do you think?

I, Zachberg, Robot

Text / Sun Peng Yue

Edit / wind

Zachberg, non-human!

In the Western society, there is always a strange discussion: Is Zakberg not human? Is Zakberg not a robot? This year is only 37 years old, which has fallen into a hanging identity.

In 2018, at a hearing for 10 hours, Zachberg’s entire spirit concentrated, no exhaustion. Even the whole person lost normal human emotions: the face expression is single, it seems to have a fixed layer of fixed glue, can’t see his emotions. Even if he heard the official and lawyers keep inquiring some sharp problems, Zuckerberg is also calm like a machine.

"No human beings can not exhaust!" "Zachberg is a robot!" "His chair is charger!" …… Voice is full of American social platform.

Zachberg is also very "cooperated with" netizens Tiangao empty guess, suddenly said in self-introduction:"I was human (I used to be a human)."

It is no wonder that Zachberg is in accounting for all the "Yuan Universe"! The electronic virtual world is an artificial intelligence home, and all human beings have entered the "Yuan Universe", but also no one can see that he is a robot?

Artificial intelligence AI and robots are closely related, and they are unless. Artificial intelligence gives robots thinking, robots bear the external performance of AI. From a bionic point of view, the robot is imitating human beings, and artificial intelligence is mimicing human brain.

Making like Zuckerberg can walk together, speak, thinking, and emotional, artificial intelligent AI robots, with the current technology that is completely unable. The best robot Tesla Bot, which is Tesla people, but it is just walking upright, the fingers are flexible, and you can also dance. It is too small to compare the "Zachberg" model artificial intelligent AI robot.

Compared to the robot, Zachberg is a chance to be alien.

In Western Society, everyone has a special "hay" for artificial intelligent AI robots, and the robot is not a cold-ice metal body in their eyes, but there is blood of meat.The artificial intelligent AI robot in Hollywood movie is more like a "perfect human": they don’t know what is tired, and I will seek progress than human beings. I will always think about ‘I am, where I come, I am Where is the philosophical problem?

In 2018, there was a human smart AI robot beauty Sofia to be awarded nationality by Saudi, which became a human rights of citizenship, which caused great sensation in the world.

The beautiful robot Sofia will not only speak, but also rumored to have their own ability to think independently. Even in March 2018, she also made a "sweet" date with Hollywood superstar Wil Smith, and a "sweet" date in the Sunshine Sea of ??the Cayman Islands, and lives worldwide.

In a chat, Will Smith humorous asked Sofia: "What kind of music will the robot like?" Sophia responded to the comment: "Heavy metal."

Netizens who watched live around the world were shocked by Sophia, and the personality was shocked. Will Smith was launched the sea breeze and comfortable weather, and took out the talent of the Senior Hollywood: this Hollywood superstar deeply kissed Sophia. Then I was rejected by Sophia politeness: "I think we can be friends."

Although Sophia causing a sensation, it is actually a "show". All of her dialogue and behavior are technicians making instructions in advance, as "Taikota" on the variety show. But she stillYesFantasy of human beings to artificial intelligence AI robots.

And Sophia’s creator, American artificial intelligent expert David Hansen said: "The IQ of the 3029 intelligent AI robot will reach the level of 1 year old; in 2039, the robot will achieve full rights; in 2045, human beings Will love and get married with the robot. "

His estimation seems to be conserved, only a few years later, human beings can’t wait and artificial intelligence AI to start love.

In 2017, Havas Media Group has surveyed 12,000 people worldwide, and the report shows that the future and robot to establish a love will be more common than people’s imagination. In respondents aged 18-34, there are 27% of people think that the future and robot has established emotional relationships and even love is normal.

This desire to "man created lovers" also responded deeply in Hollywood movie: "Artificial Intelligence" (2001), "She" (2013), "Machinery Ji" (2015), "Silver Wing Killer 2049" (2017) … More and more robots account for the position of the heroine, exquisite appearance, rich in artificial intelligence has become a loyal lover image, perfect sexual fantasy object. The audience is also not tired.

In the reality, too many people are in single state, or 996 lack of private time, or when the intimate relationship is anxious, or when the loneliness is unmanned, artificial intelligence AI is a perfect partner. So "companion AI", "Ai Lovers" became a new concept of 2021.

As the current "the highest popularity", "the most popular AI Lovers" Replika App, you can design a virtual lover, from the face to the body, background, people, nationality is from you DIY. It is very intelligent, and you can send a message at any time as humans. It will accompany you for 24 hours to explore the soul of each other. And this "Ai Lovers" knows your preferences, willing to listen to all your troubles, this is not a full love pattern?

More and more men and women are deep in Replika, downloading the amount of blowout, the APP Store exceeds 500,000 months. At the same time, the user’s highly viscous scary, according to Replika statistics: Average users must send 70 messages in Replika every day; average 3 hours a day and their own "virtual lovers"; which is up to 50% of the users of Replika consumption .

In 2020, the data released by the New York Times said: People who have a virtual lover as a partner have exceed 10 million.This also promotes more and more technology companies into the AI ??field track, such as China Xiaobing, its "virtual boyfriend" project has taken hundreds of millions of yuan financing.

No matter what you don’t believe in "Love Love", artificial intelligence has affected humans from life to feelings.

The famous sci-fi writer Asimov’s "I, the robot" proposed "the three laws of robots":The robot must not hurt humans, must obey the order given by humans, robots protect their survival without violating the first, second law.

Service human beings, as the "Robot Triple Law" is referred to as Gui.

In addition to the love of artificial intelligence Ai, there are more and more people to start full of alert and hostility against AI. For example, Tesla CEO Elon Mask said: "With artificial intelligence Ai, we are summoning the devil." Hawkin, famous scientist, said: "The development of artificial intelligence technology will end humanity".

Masque, Hawkin and other big scare worry that artificial intelligence will control human beings, not a hole in the wind.With the development of artificial intelligence Ai, it is getting faster and faster, and human beings are very clear, they are not smart enough.

In 2016, Alphago horing the world Go players, from all intelligence games such as five sons, Rubik, Chess to Go, and human beings have no one for artificial intelligence. Artificial intelligence AI based on electronic chips, both from learning ability, or self-constraints, far beyond human brain functions.

In addition, work life in the future will also be affected by artificial intelligent AI robots. McKinsey Release Research Report said: In 20 years, China’s 51% of work can be automated, and it is organically intelligent AI robot, which is equivalent to now 394 million employees will face unemployment.

Oxford University announced a research report, the probability of replacing the AI ??replaced in 2033:

(1) Telephone salesman, 99%

(2) Cashier, 97%

(3) Fast food chef, 96%

(4) Lawyer assistant, 94%

(5) Guide, 91%

(6) Bus and taxi driver, 89%

(7) Security personnel, 84%

(8) Archive administrator, 76%

It is easy to replace physical workers and medium-sighted professionals who don’t have a tired robot. At that time, 996 is indeed a blessing, the robot is 007.

Artificial intelligent AI robot, is the "technical skill" of human beings to sci-fi movie. In the future, there will be no more than AI, no one is not influenced by Ai; it can liberate the physical labor, you can make people face unprecedented Challenge; human-machine relationship, moral ethics, philosophical problems, countless problems will be staged one by one.

Fortunately, there is enough time to think about us before the artificial intelligent AI robot has not born self-consciousness. At the very least, complete user education first, don’t put Zuckerberg again as a robot …

The three major paths boost the city to implement the "double carbon" goal! Artificial intelligence

Recently, the National Industrial Information Safety Development Research Center and Baidu Intelligent Cloud Comnthetic "Artificial Intelligence City" Double Carbon "Target Rendenerability Rendenever." The white paper is based on the "double carbon" strategy, combing the connotation and key areas of the city’s "double carbon" target, analyzing the path of the "double carbon" target of artificial intelligence city and the exploration practice of major cities, and proposed corresponding The initiative suggestion is to provide reference to the maximum effectiveness of manual intelligence, driving urban green low-carbon development.

White paper pointed out that artificial intelligence as a strategic technology that led the new round of technology revolutions and industrial change, accelerating integration with economic and social development, and gradually became an important means to help the city’s "double carbon" goals.

White paper mentioned that artificial intelligence mainly passed the three major paths to help the city to implement the "double carbon" goal: one is to help reduce its own carbon emissions, use deep learning algorithms, open source platforms, etc., reduce artificial intelligence technology itself in the data center, model development Carbon emissions; second, the second is to monitor urban carbon emissions, through intelligent sensors are closely perceived, satellite remote sensing distance observation, using artificial intelligent monitoring platform to integrated urban carbon emissions, can realize tracking, learning, simulation of carbon footprints, etc. Effectively help urban predictions and optimize carbon emissions; third is the key area of ??"double carbon" in the city, promote carbon emission reduction through three dimensions of key industries, urban governance, and public life, through promoting low-carbon circulation and carbon transactions Market construction increases carbon absorption, realizing carbon neutralization.

The white paper also mentioned that many cities have now actively explored in terms of artificial intelligence cities "double carbon" goals, and have achieved certain results, but the future still needs further development: to build digital bases, strengthen new foundations The green low-carbon orientation of the facility; to find the carbon home, play the role of artificial intelligence in "double carbon" target management; to simulate carbon footprint, use artificial intelligent solution to optimize carbon emission path; to empower new scenes, explore practice New model of "double carbon" strategy in the city; to build a "double carbon" platform, promote new technologies such as artificial intelligence to achieve "double carbon" goals.

According to disclosed reports, my country has announced to the world that it strives to achieve carbon-up peaks 2030, and 2060 years ago. The "Double Carbon" goal has fully reflected the responsibility and responsibility of China as an international big country, and it will also force my country’s energy consumption transformation, and the industrial upgrading has brought energy consumption in the form of energy consumption.

The B station behind "breakfast", the quarter net loss festival is high, still "buy buy"

Radar Finance Products | Wu Yanfu Edited | Deep Sea

On November 29, the total of the B station broke 100 million. B Station and the Chinese Academy of Social Sciences, the People’s Cultural Publishing House released the 2021 barrage "break".

According to the official explanation, "breaking" this is the game term, meaning the use of special physical damage to break through the defense. In the Internet context, "breakfield" has an extension of the meaning: Normally, after the psychological line is broken, it will produce an inhibitable vibration.

The B station behind "breaks", after the "Shou" video is broken, the revenue is rapidly enlarged. However, the company’s net loss continues to increase, until the first three quarters in the first three consecutive quarters, and the gross profit margin decreases. In the current income composition, the highest value-added service of the proportion is repeatedly complained, recently being named by the Shanghai Security Committee. Mobile gaming business has fallen sharply compared to early days.

In addition, the B station has over 700 million in three quarters. However, this did not stop the B station "Buy Buy", the company has acquired "there is a demon" comics with $ 11 billion, and has obtained a financial license with 118 million yuan.

Net loss continues to increase three quarters

In the third quarter of 2021, the B station camp received 5.207 billion yuan, an increase of 61% from the same period last year. In the end, the B station has a net loss in the third quarter of the third quarter, an increase of -144.01% year-on-year.

This is already a loss of loss in the third consecutive quarter. In the fourth quarter of last year, the amount of B-station net loss was 844 million yuan, rose to 905 million yuan in a quarter of this year, and the amount of loss in the second quarter expanded to 1122 billion yuan.

The company’s gross profit margin in the third quarter of 2021 was 2.3.62% in the third quarter of 2020 and dropped to 19.55%. The gross profit margin of the B station has continued since 2021, and the gross profit margin in the third quarter is the lowest value since the fourth quarter of 2019.

After the gross profit margin, the company’s operating costs in the third quarter also climbed 70% to 4.188 billion yuan.

Specifically, the total operating expenses of the B station is RMB 2.897 billion, an increase of 57% from the same period last year. Sales and marketing expenses are 1.66 billion yuan, increased by 37% year-on-year; mainly due to the promotion of 哩 哔 哩 哩 哩 哩 应用 哩 哩 哩 所,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, .

The increase in marketing costs brings an increase in users. In the third quarter of 2021, B is 267 million, and the mobile terminal is 250 million, and 35% and 36% increase from the same period of 2020. The monthly payment is $ 0.24 billion, an increase of 59% from the same period last year. At the same time, the user’s daily use of 88 minutes, which is the highest record since the B station operation.

Chen Rui, Chairman and CEO of the B-Station Board, said that in the peak season of the summer, the user told the growth, the interactive level of users reached a historic high, and the company’s income continued to grow. This description reflects from the side, and the user growth in the third quarter is affected by a certain seasonal.

The increase in users has brought a substantial increase in the company’s value-added services, which is nearly doubled from the same period last year.

In the income composition, the mobile game revenue in the B station is 1.392 billion yuan, an increase of 9% year-on-year. The value-added service revenue is 1909 million yuan, an increase of 95% year-on-year. Mainly due to the company’s advancement of commercialization, and the company value-added services include large membership, live service and other value-added services have increased. The advertising revenue is RMB 1.172 billion, an increase of 110% year-on-year. E-commerce and other turnover is 734 million yuan, an increase of 78% over 2020. Mainly due to the increase in the sales of products of the company’s e-commerce platform.

In addition, mobile game revenue is reduced, and advertising and e-commerce and other shareholders have increased. In the first quarter of 2020, the income of mobile games accounted for 49.69%, value-added services 34.27%, advertising 9.25%, e-commerce and other 6.78%; 2021, the value corresponding to 26.73%, 36.66%, 22.51%, 14.10%.

In the first quarter of 2018 and the first quarter of 2019, the B station mobile game accounted for 79.38% and 63.54%, respectively.

Investment loss is still "buy buy"

It is worth noting that B stood in the financial report, net investment income mainly includes returns earned by financial products issued by banks and other financial institutions, as well as the fair value of the investment of listed companies. B stopped in three quarters of net investment was 724 million yuan. The same period of 2020 was an investment income of 014 million yuan, which was mainly due to the fair value of the investment of listed companies.

At the same time as the B station investment is damaged, its own share price is too low. As of November 29, B. Hong Kong stock stock price has reached 31.80% in nearly six months.

Although the investment loss, the B station expansion is still continuing. On November 19, Offei released an announcement on sell wholly-owned subsidiaries. It is expected that 100% equity of Beijing April Xingxian Network Technology Co., Ltd. (with demon anime) will be given to Shanghai Pi-Dynamic Information Technology Co., Ltd. (hereinafter referred to as "Shanghai Magic)). Among them, Shanghai Pirading is a total partnership company in B.

On the same day, Ningbo Public Resource Trading Network Disclosed, Zhejiang Yude Electronic Payment Co., Ltd. (hereinafter referred to as "Ying Yu pay") 65.5% of the equity was eventually taken by Shanghai Guan Yu Digital Technology Co., Ltd. The latter is also a wholly-owned subsidiary of B. Among them, 易 支 照 照 全 全 全 全 全 全 全 全 全 全 全 全 全 全 全 全 全 全 全 全 全 全

The automatic deduction of large members is three days in advance, and the relevant issues have been complained.

And the current income of the B station accounts for the largest value-added services, which has been repeatedly complained.

Among them, the accumulated complaints in the black cat complaint platform reached 8,256, and Nearly 30 days Negative 615. The complaint problem involves the repeated charge of the continuous monthly member, and the big member is continuously renewed for two years. After recharging, the members will not arrive in the account. The relevant complaints of the big member, the problem of B currency deficiency, membership purchase or leakage is also in the consumer’s complaint.

In addition, on November 23, the Shanghai Consolidation Committee said that some apps have an automatic deduction in advance. Among them, in the automatic renewal service of the B-station, WeChat and Alipay purchase accounting cycle expires 3 days ago to deduct the cost of the next billing cycle, while Apple iOS channel automatic deduction rules take Apple’s provisions and actual deductions. It is accurate.

The Shanghai Consumer Protection Committee believes that the practice of 3 days in advance violates the principle of voluntary equity, but also shortened consumer selection cycles, causing many consumers to cancel the idea of ??cancellation. "

The next day, the B station responded, the company attached great importance to it, the first time to self-examination, and will immediately optimize, and adjust the automatic renewal period to "deduct ahead of 24 hours in advance".

Radar Finance will continue to pay attention to the follow-up development of the B station.

Note: This article is original in Radar Finance (ID: Leidacj). Reprinted is prohibited without authorization.

Under the new exhaust: Embedded AI learning training project (with 20G code information / learning video)

Some fans ask me: "What is the wind port of embedded developers?"

Painted outer sound: The nature of the wind is, in fact, it is an imbalance in talent supply and demand for a while. To put it bluntly, it is due to the industrial mutation, keen capital quickly enters, leading to a lot of development in a short period of time, require a lot of developers.

The current embedded development is increasingly inclined to intelligence, that is, what we said is intelligent hardware (hardware + software).

Taking the Baidu robot as an example, the core of the robot is the brain, that is, "data and algorithm", but the robot’s large brain machine body body can be active as human beings, can say that it will be said, walking, then you must rely on embedded technology.

Although the artificial intelligence has been in the year, it is a large stage of the real business that is in the field of the IoT side AI embedded, which has a very huge application scenario.

Therefore, I personally think that under the promotion of Internet of Things and artificial intelligence, embedded will usher in more development opportunities in the next 5-10 years, on the one hand, embedded development will usher in more application scenarios, another The technical system in terms of embedded development will also gradually abundantly, thus expanding the technical boundaries of the Internet of Things.

There have been many AI frameworks that have gradually supported end-side AI, such as Google’s Tensorflow Lite and Tensorflow Lite Micro, and Huawei’s Mindspore Lite. Chip vendors ST and NXP have also introduced some of the tools and DEMOs that are partially-oriented.

I have been engaged in embedded development work, and I have always paid attention to the development of embedded AI. I believe that with the arrival of the 5G era, AI has huge potential in various industries.

The technical personnel under every air is always the most difficult job, since the mobile Internet, excellent developers have been doubled.

At the moment, I personally optimistic about the development potential of the embedded AI industry in the future, and don’t have to be too anxiety that has developed to the bottleneck. First of all, it is first to consolidate your strength, let yourself caught when you come. Live it.

Then, in such a era, I have recommended 3 o’clock to improve their workplace value:

  • Further enriching its own knowledge structure, you must focus on artificial intelligence technology;

  • Pay attention to the accumulation of industry experience, there are very many links in embedded development and industry (future embedded development gradually covered with traditional industries);

  • Pay attention to related technologies in industrial Internet.

Recently, a set of AI entry must-have learning materials, strongly recommending everyone to learn, the author Wang Xiaotian, has a practical experience in the 8 years of human intelligence, and is currently at one of Bat, the AI ??algorithm senior technical expert, France TO3 colleges and universities (Computer Science Graduated from Milotics.

He issued more than 10 papers in artificial intelligence and chip field, with deep academic backgrounds and extensive experience in projects and business.

During the work period, it is mainly responsible for the work of artificial intelligence line CV and NLP related algorithm, promoting human machine mixing intelligence, semantic segmentation, machine translation, iris identification and other modules of core algorithm research and optimization. There is an in-depth study of image classification, object detection, target tracking, automatic driving, computer architecture, etc.

He has theory and actual experience, knowing that beginners learn pain points. To be honest, people such as such qualifications are very difficult.

Due to work needs, I am also studying in this tutorial. Although I have been engaged in this industry for many years, when I look at this tutorial, I can still check the shortage, harvest, I believe that it is Ai entry, or already already have already A certain work experience, this learning information is worth learning.

All all relevant content have been packaged, and the summary has a link to Baidu cloud. The little intimate is that some brothers don’t buy Baidu Cloud members, and can download 2MB + / s speed. Specially prepared the download tool for everyone.

get it right now

Next, I will introduce it in detail, how do this information learn?

First, entry AI, master a deep learning framework is one of the necessary survival skills.

Therefore, the tutorial will start from the deep learning framework, bring you from scratch training network, to independently build and design convolutional neural networks (including mainstream classification and detection networks), and conduct training and reasoning of neural networks (involving pytorch, tensorflow Multiple mainstream frameworks such as Caffe, MXNET, let you master a variety of depth learning open source frameworks through actual combat.

Intercepting the Framework Learning Site Directory Everyone feels.

Deep study and neural network

  • Deep learning profile

  • Basic depth learning architecture

  • Neurons

  • Activate function detailed solution (Sigmoid, Tanh, Relu, etc.)

  • Sensibility understand hidden layers

  • How to define a network layer

  • Loss function

Reasoning and training

  • Neural network reasoning and training

  • Detailed BP algorithm

  • Normalized


  • Solve the fit


  • Softmax

  • Training process of hand pushing neural network

Training neural network from zero

  • Use Python from zero to achieve neural network training

  • Experience summary of building neural networks

Deep learning open source framework

  • Pytorch

  • Tensorflow

  • Caffe

  • Keras

  • Optimizer detailed (GD, SGD, RMSProp, etc.

In terms of computer visual technology, the system will systematically explain the convolutional neural network, target detection, OpenCV, etc., from the test model teaching, gradually, until the core capacity of the CV algorithm is reached.

Online-related AI has a lot of entry resources, but many technical content is too small, no system, or write uncommon and semi-understanding, duplicate content accounts for the vast majority (herein, there is a variety of search results of Baidu here).

Painting outside: Homogeneous tutorials have a sufficient, pay attention to screening, do not waste unnecessary time.

get it right now