Takeaways from Tesla's AI Day

Takeaways from Tesla's AI Day

Tesla is the world's leading electric vehicle manufacturer, with a 15.2% market share - according to the first half of 2021. However, Tesla is not just an EV manufacturer. It is an AI company that manufactures electric vehicles.

On Thursday evening, Tesla staged an event in Palo Alto, California, to demonstrate the brand's autonomous vehicle and artificial intelligence technology. The AI Day event was similar to Tesla's Battery Day last fall in idea, though this AI event was likely more focused on hiring new engineers than impressing investors and the general public.

Tesla demonstrated its self-driving software in action, made bold claims about future advancements to the technology, and vowed to construct a prototype of a humanoid robot within the next year during the event, which can be viewed on Tesla's YouTube channel. All of this comes amid a fresh investigation by the National Highway Traffic Safety Administration (NHTSA) into 11 incidents in which Teslas collided with parked emergency vehicles while using Autopilot or Traffic Aware Cruise Control.

Tesla's Neural Net Planner is able to predict the trajectory of objects

The day began with a video depicting a Tesla managing traffic, intersections, and other obstacles without the assistance of a human driver. Following that, Tesla's AI researchers went into great depth about how Tesla's computers interpret data from the self-driving system's cameras. Tesla has chosen to use exclusively information from cameras to inform its cars' driver assistance functions, which has sparked some controversy. Cameras are often used in conjunction with radar or lidar sensors in many driver aid systems.

Elon Musk, the Technoking of Tesla, stated that the existing technology in Teslas is "at least 200% to 300% better" than a human driver, and that a new hardware setup will be even more adept. After its production timeline was quietly pushed back earlier this month, Musk said the new hardware will appear in the Cybertruck "in a year or two," giving us our first idea of when Tesla's pickup might hit the roads.

Tesla's supercomputer, Dojo, and the proprietary microchip Tesla created to run it will be responsible for all of this capability. Tesla created its own microprocessor in part to sidestep the supply-chain issues that have led automakers to halt new car manufacturing on a regular basis over the last year. Tesla engineers had to rebuild their own software to work on the alternate microchips Tesla sourced when its prefered chips were unavailable, Musk revealed during an earnings call earlier this year.

The D1 chip, which will power Tesla's Dojo supercomputer, was unveiled. It's also projected to be the world's most powerful AI training computer.

The Dojo training tile is composed of tiled D1 compute modules.

Why is does this matter? Massive volumes of data and a strong computer capable of using that data to train deep neural nets are two requirements for artificial intelligence. Tesla already has an advantage over other automakers, with over 1 million autopilot-enabled EVs on the road.

Tesla's advantage has now been bolstered by the introduction of an exascale supercomputer, which management claims will be operational next year. In other words, the business has more data and better technology than the competition, putting it leagues ahead in the race to produce a self-driving car. And autonomous vehicles are the way of the future in the automobile business.

To conclude the event, Elon Musk took the stage with a dancer dressed as a robot and said that Tesla is developing a humanoid robot prototype capable of deadlifting 150 pounds and running at 5 mph. Musk stated that the robot will be benign, but that it was built such that humans would be able to outrun and overpower it. Musk may be in good shape to outrun a super robot, but we would probably feel more comfortable having a robot that can't run at all.