小印跪舔翼龙: 肉麻慎入
版主: 牛河梁
#9 Re: 小印跪舔翼龙: 肉麻慎入
这个小印是负责自动驾驶研发的。看几个FSD (Supervised) v12.4.1的视频
持仓抄底锁利,你钱你定
看牛观猪喊熊,自娱自乐
股市变幻莫测,不作不死
赌途曲折无常,吃枣药丸
看牛观猪喊熊,自娱自乐
股市变幻莫测,不作不死
赌途曲折无常,吃枣药丸
#10 Re: 小印跪舔翼龙: 肉麻慎入
小印的推特全文
Elon and AI at Tesla
- by Ashok Elluswamy
Elon Musk @elonmusk has been the key driver of AI and autonomy at Tesla. He has always pushed us to achieve great things, even when such ideas were seemingly impossible at the time. Some examples:
- Back in 2014, Autopilot started on a ridiculously tiny computer that only had ~384 KB of memory and puny compute (didn't even have native floating point arithmetic). He asked the engineering team to implement lane keeping, lane changing, longitudinal control for vehicles, curvature, etc. Many, even in the team, thought that the request was crazy. Nonetheless, he never gave up and pushed the team to achieve this very difficult goal. In 2015, beyond all odds, Tesla shipped the world's first Autopilot system. The second closest such product only came to market many years later.
- In 2016, Tesla started doing all of the computer vision required for Autopilot in-house instead of depending on external vendors. Many people thought it was insane to bet the product on developing the vision system from scratch within a few months, which had taken other companies a decade or more. Yet, we achieved this target within eleven months. This was a strategically important move that started the development of a strong AI team at Tesla.
- Not only did he push for strong AI software, but also for powerful AI hardware. Tesla, which others thought was just a car company, was making custom silicon to run neural networks efficiently. This hardware that was originally designed in 2017, came to production in February 2019 and remains extremely competitive with hardware coming out to date. For reference, this five year old AI computer has roughly 8x the AI inference compute as the state-of-the-art Apple M3 chip. It is still able to run the latest end-to-end neural networks built on top of the latest AI technology.
- He was the one who bet on vision and AI to solve autonomy instead of relying on sensor crutches and high-definition maps. For anyone who has experienced the latest versions of FSD, it might be obvious that it can see all the important things and drive the car based on pure vision. However, back in 2020 and earlier it wasn't obvious to most. In fact, many "experts" in the field ridiculed Tesla and Elon for these choices. We have proved them wrong by shipping supervised FSD to millions of cars and shown that with good AI software, the car is able to handle the complexities of city driving such as making turns, handling intersection, yielding to pedestrians etc., just by seeing outside. In fact, we even removed the radars and ultrasonics to just really focus on the heart of the problem, which is AI. Today, it's almost paradoxical that, Teslas have the least amount of raw sensors, yet have the most autonomous capability compared to any production car. Pulling off such a contrary bet was only possible because of his extreme conviction and deep understanding of this problem.
- He kickstarted the work on humanoid robots at Tesla in 2021, again before any ChatGPT or other obvious examples of the rise of AI. Just like the vehicle autonomy, Optimus is also being developed to be competent, scalable, and cost-effective in order to widely serve the world.
持仓抄底锁利,你钱你定
看牛观猪喊熊,自娱自乐
股市变幻莫测,不作不死
赌途曲折无常,吃枣药丸
看牛观猪喊熊,自娱自乐
股市变幻莫测,不作不死
赌途曲折无常,吃枣药丸




