Tesla, Inc. AI Day will be held on August 19th. Musk has previously said that the AI Day will showcase Tesla, Inc. 's progress in hardware and software development in artificial intelligence and recruit top talent.

As the day of the event is approaching, Goldman Sachs Group predicts what kind of "big move" Tesla, Inc. will release in his latest research report released on Aug. 16.Goldman Sachs Group expects Tesla, Inc. to discuss five major contents, namely:AINeural network, FSDSoftware, self-developed supercomputer Dojo, the next version of Hardware 4.0And the application of artificial intelligence outside the field of autopilot.

AI neural network

First of all, Goldman Sachs Group expects Tesla, Inc. to discuss the AI neural network and how it is trained at the event.

Tesla, Inc. 's neural network supports its Autopilot/FSD (fully autopilot, Full Self Driving), which mainly deals with key data needed for driving functions such as object recognition and road planning. This neural network is functionally and structurally similar to the primary visual cortex of the human brain, allowing the vehicle / program to identify the surrounding environment and determine what the vehicle needs to do based on big data.

Goldman Sachs Group said Tesla, Inc. believes that the key to successfully training his neural network is a large number of, diversified and real-world databases.And this is collected through its huge fleet. Goldman Sachs Group estimates that the current number of fleets in the world exceeds 1.5 million.

Most importantly, even if AutopilotWithout startup, Tesla, Inc. can also be in "shadow mode" (shadow mode)) to simulate and test the driving decision of neural network.This allows Tesla, Inc. to identify the differences between neural networks and human driving decisions, and further improve his FSD software.

Tesla, Inc. mentioned the so-called "shadow model" in a report submitted to the California vehicle Administration (Californian DMV) in early 2018. Tesla, Inc. said at the time that the company did not test fully self-driving cars on California roads in 2017, but the cars tested self-driving technology in "shadow mode" during normal driving.

FSD software

In addition, Goldman Sachs Group also expects Tesla, Inc. to discuss the advantages of FSD software, including its purely visual approach to artificial intelligence training.

At present, the latest beta version of Tesla, Inc. FSD software Version 9 (V9) is V9.2 launched through the OTA aerial software update in July 2021.

V9 is a vision-specific L2ax 2 + system. The term "visual dedicated" here refers to the use of surrounding cameras to collect data, rather than radar or lidar, while the L2Accord 2 + system means that the vehicle can drive itself in many cases, but the driver needs to be ready to control the vehicle at all times.

Goldman Sachs Group believes that Tesla, Inc. 's implementation is based on visual FSD.A key aspect of is the ability to use large amounts of data to train neural networks, and the ability to process 3DVideo /A marking tool for multiple images.

Self-developed supercomputer Dojo

Goldman Sachs Group also expects that Tesla, Inc. will discuss the progress of self-developed supercomputer Dojo on the event day.

In fact, Musk has mentioned Dojo since at least 2019. Musk has said that Dojo will be able to handle huge amounts of video data to achieve pure visual autopilot.

At the CVPR 2021 workshop held in June this year, Tesla, Inc. AI director Andrej Karpathy introduced the latest situation of this supercomputer: with 10petabytes NVME storage and running speed of 1.6 terrabytes; per second, it is finally 1.8 EFLOPS per second, which may become the fifth largest supercomputer in the world.

The next version of autopilot hardware Hardware 4.0

Goldman Sachs Group said that Tesla, Inc. may also discuss the next version of autopilot hardware Hardware 4.0 at the conference.The progress of.

Tesla, Inc. discussed his proprietary FSD hardware Hardware 3.0 on Autopilot Day (Autonomy Day) in 2019. The project began to open in 2016 and began to be carried on Tesla, Inc. cars in 2019.

For Tesla, Inc., the internally designed computer provides better performance, complete redundancy of key driving functions, and lower cost.

Application of artificial Intelligence outside the Field of Autopilot

At the end of the article, Goldman Sachs Group boldly predicted that Tesla, Inc. would discuss the application of artificial intelligence outside the field of autopilot, including solar energy /Storage software, advanced manufacturing in Tesla, Inc. factory, etc.

Goldman Sachs Group said, for example, Tesla, Inc. 's energy software Opticaster is a general platform for driving Powerwall, Powerpack and Megapack. The platform uses machine learning, prediction, optimization and real-time control algorithms to reduce energy consumption costs, increase renewable energy consumption and realize microgrid control.

In addition, Tesla, Inc. has always stressed that manufacturing is its core competitiveness and long-term competitive advantage. Goldman Sachs Group believes that Tesla will discuss how advanced robots and AI in factory systems can improve efficiency, reduce costs and increase output.