Very Interesting Data On Tesla FSD

⭐️ Very Interesting Data On Tesla FSD

Until V11, there was a high frequency of situations in which people from a specific area had to intervene,
In V12, the frequency of human intervention in the same area converges to 0% (the population will be very small, of course)

The places that V11 users said were never successful in the past three years
Most of the users say it’s all possible in V12
In other words, the issue that was impossible to drive due to the form of a road or signal system
Approaching a resolution step in V12
There are significant improvements to the extent that there are intermittent interventions in certain situations and unexpected situations

We’re now in the United States and Canada, and we’re in all Tesla vehicles
One-month trial begins to be distributed (even for vehicles that have not purchased FSD)
Canadian, U.S. Users’ Overwhelming Positive Ratings

I only experienced self-driving that I can’t go back to in Korea
I think it’s very difficult to evaluate it because it’s self-driving
At least go to SF and ride Waymo
I think I need to experience the version of FSDV12.3.2 or later in the U.S. and talk about it
I’m thinking of renting a car through Turo when I travel to the U.S

Of course, I hope that FSD will be introduced in Korea before that.

✅ FSD is activated by navigating in the parking lot and pressing the handle button -> Automatic parking at an empty spot on the side of the road arriving at the destination with FSD -> Automatic parking is possible by pressing the button with the vision auto park -> After going to school and pressing where to park in 2 months, the car will go and self-park if you get off

In 6 months, North America will be able to leave and arrive at the destination with a finger button click of two
While using the service, we collect additional data from the edge case
We will mass-produce the robo-taxi platform when the $25,000 vehicle comes out

✔️ AI-based automatic parking video on the video I shared
✔️ Video leaving with only the FSD button pressed
✔ There is a video of self-parking on the road when arriving at the destination of the ️ Navy

🔥For reference, Tesla’s FSD moment will be available for free for all 1.7 million Tesla users in North America for a month starting next week

The chage was amazing. Technically it wasn’t that amazing compared to the GPT-3, and serviceally it was only slightly different packaging from Playground etc, but the most surprising thing was that the world was amazing.

I’ve been watching deep learning technologies/services for a long time, and I’ve been feeling that the world is not very surprised even if these technologies/services come out. Then, the world’s reaction to Chage shocked me to the point where I wondered if it was possible to say that the world changes in one morning.

The same goes for Tesla’s FSD. How can the world still be surprised when they show such amazing driving skills? Will they say that other companies are still at a low level or that they can catch up quickly? Perhaps the Chage moment will come soon.

Since the recent launch of the FSD beta, there seems to be a lot of controversy between Tesla and the LiDAR camp, and in conclusion, I think the winning streak has leaned a lot toward Tesla.

As is well known, the self-driving methods between the two camps are different. In the case of Tesla, an autonomous driving system was established only with pure computer vision sensing without LiDAR and radar, while Alphabet’s Waymo, a representative LiDAR camp, operates a system that integrates three LiDAR-Camera. Both Tesla and LiDAR camps have their own strengths and weaknesses, but I think the fundamental difference that will determine the outcome lies in ‘expandability’.

1. LiDAR camp with clear scalability limits

First of all, in the case of LiDAR camp, stable autonomous driving is possible only when there is a high-resolution map through hd mapping of topographic features in a specific area in advance. This is why Waymo travels only within 10% of the city of Phoenix, Arizona. In addition, in the case of LiDAR, it is necessary to update and provide high-capacity and high-definition LiDAR maps in real time, but not only is it difficult to update the topographic features in real time, but it also incurs tremendous traffic and cost, which reduces power efficiency. The power consumed by LiDAR driving is also significant. In addition, there are limitations in obtaining and building such detailed map information, especially outside the United States, and in the end, it is concluded that it is not easy to provide services to the entire world. It reduces the competitiveness of the expensive LiDAR itself and its cumbersome visual.
What’s interesting is that Waymo’s corporate value, which was priced at 200 trillion won just two or three years ago, shrank to the level of 30 trillion won, and Waymo CEO John Krafcik resigned, taking responsibility for the huge deficit and lack of performance. In addition, for the first time, Waymo began receiving additional funding only with external funds without internal Alphabet funding, and I think this has led to the judgment that even Alphabet can no longer continue to pour water into the bottom of the jar. In addition, the market capitalization of Velodyne LiDAR, a leading LiDAR company, has also been cut by one-third in about half a year (1.8 trillion won). There are many things to say about not being a LiDAR in the world, but once you look at the current market evaluation, the future of LiDAR does not seem very bright.

2. Tesla’s Universal Self-Driving Technology Open to Scalability

Tesla, on the other hand, is capable of driving without much difference in performance even when thrown on roads in random cities around the world because it inputs and determines the surrounding features in real-time using only eight cameras. This is why Tesla focuses on maximum computational power (98%) in image acquisition technology.
As a result, the advantage of this method is that subscription services will be available in all regions of the world as long as the completion of autonomous driving is improved in the future.
Not to mention the amount of driving data compared to Waymo (250x difference), it is also impossible to compare in terms of quality (Real World Data vs Tested Data).
Tesla’s recent launch of a pure camera vision FSD, except for the Radar, which functioned as an auxiliary device during the Autopilot era, is due to the fact that the accuracy of the vision has improved to more than radar, and if the judgment of each camera and radar conflicts with each other, radar rather makes noise and interferes with accurate judgment. The advantage of LiDAR is distance recognition, and Tesla is overcoming this by implementing pseudo-Lidar technology with only cameras

3. Tesla, the World’s Top AI and Semiconductor Company

For meaningful AI technology to be completed, three elements are needed: data, algorithm, and supercomputing. Tesla is the only automobile company that vertically integrates all of these, with more than 1 million vehicles worldwide, including actual road data (more than 4.8 billion kilometers cumulative), top-notch software and top-performance supercomputers. (There are no big tech companies among all companies.)
In addition, it has recruited a super genius in the chip industry called ‘Jim Keller’ and is designing semiconductors that are optimized only for Tesla’s autonomous driving without outsourcing. Through this, data processing power and energy efficiency are maximized, which eventually leads to an increase in mileage.
However, other automakers have no choice but to cooperate with general-purpose semiconductor companies such as Nvidia due to lack of chip design capabilities, and Nvidia chips are not designed for specific companies’ vehicles, but for each car maker to produce optimized performance like Tesla, and there are many errors and power efficiency. Tesla’s self-driving chips are already six years ahead of Volkswagen and Toyota, and three years ahead of Nvidia.

4. Software Beyond the Human Realms Tesla in the 2.0 Era

Tesla’s self-driving performance improvement rate seems to have recently risen from “arithmetic exponential” to “geometric exponential,” as the software 2.0 era has opened.
In the era of Software 1.0, it was necessary to “label” all kinds of objects in order to recognize and judge objects, which surprisingly was accomplished by changing the labor force of many people. This is because the accumulation of numerous inductive data was first required to lay the foundation for an early stage of consistent judgment without errors. However, this was clearly a limitation in the autonomous driving area, which had to pursue near-perfect safety in a real world with near-infinite exceptions.
However, another genius in charge of Tesla’s AI, Andre Kafasi, recently declared the “Software 2.0 Era,” and as a result, supercomputers enter sufficient data without humans having to write algorithms, and they have entered the stage of developing and improving logical structures and algorithms on their own through artificial neural network learning.
Tesla is set to launch Dojo, the world’s fifth-highest performing supercomputer, later this year, and much of it is expected to be revealed at this AI Day.

Collecting driving data and sending headquarters -> Labeling supercomputers, self-development of AI through deep learning -> updating Tesla lane software around the world through OTA -> collecting better driving data -> repeating indefinitely….

Creating this virtuous cycle repetition structure, personally, I honestly think the game is almost over. This is because it is already becoming almost impossible for other companies to overtake Tesla in the quantity and quality of driving data. According to Professor Andrew Ng, one of the world’s top AI authorities, data determines 80% of deep learning success. As in history, it is difficult for a latecomer to overturn a leader in the data platform market.

5. The one who controls the OS rules the world

During the era of PCs, Microsoft dominated the world with the OS called Windows, and in the era of smartphones in recent decades, Apple on iOS and Google on Android have dominated the world. After all, those who dominate the device’s OS that penetrates the paradigm of the times have monopolized the added value through it.
We are about to face another paradigm shift in which key devices that dominate the production and consumption of daily data are transferred from “smartphones” to “computers on wheels.” Who will be the owner of the OS that dominates this new era?
Even for criminals like me, the answer seems obvious, but many people still think differently. Still, with a long breath, it’s not small for Tesla investment