Posted

Last Updated:

by

in

, ,

Andrej Karparthy talks Tesla full self driving, Optimus and AI in No Priors podcast

September 14, 2024

The No Priors podcast episode 80 released September 5, 2024 titled “The Road to Autonomous Intelligence with Andrej Karparthy”, talked about Tesla full self driving, Optimus and AI. https://www.youtube.com/watch?v=hM_h0UA7upI

Andrej says Waymo’s full self driving took 10 years to go from working demo to production. According to Andrej, Waymo’s solution uses lots of sensor and will have a tough time scaling. Andrej says Tesla still has issues to work out on the software but he argues that software is an easier problem to solve and thus gives Tesla the advantage over Waymo. Tesla has the hardware and scale and this is a harder problem. He says that Tesla like Waymo also uses expensive sensors like lidar besides cameras. The difference is Telsa uses those during the training phase. In test or deployment phase, it is cameras only as he believes the information required for driving is in the pixel.

Andrej was asked how much of full self driving transfers to Optimus robot project. Andrej sees a lot that can be transferred over. The robot is driving in walkable space. The humanoid form is important in that it makes tele operation training easier. He says Tesla is able to make quick progress on Optimus because the company already have a lot of the equipment, tooling and expertise.

On AI, Andrej explains that the transformer model is very generic neutral network that’s capable of solving any function. This model hasn’t reached its limit yet and researcher are advancing it by tuning cost function and the quality of the dataset. The entirety of the internet has a lot of junk but there is also traces of human intelligence and thinking traces. By removing the junk, he believes models in the future can be a lot smaller.

Towards the end, when asked what young kids should be studying, Andrej recommends computer science, physics, and math. He believes they help you think. Thinking skill is important before AI and he believes it will continue to be important with AI.