
Chubster
12.6K posts



I don’t think people realize how affordable Teslas have gotten. A new Model Y is $530/month. (72 months)


Teslas are in tunnels under Vegas transporting millions of people







This 93 year old has found new freedom after she bought a new @Tesla Model Y with FSD. She also uses Grok navigation. "Although she has always been a good driver, my mom can now drive without the fear or fatigue that can naturally come with age. No more relying on others for every trip. No more feeling stuck. This is true mobility that can spark new adventures in a still adventurous women!" (via Dan Doyle's Family Channel. Full video below)



Falcon Heavy is so beautiful


Flexing a win today 💪🏼 Ags nearly had that series, too.







FSD Supervised isn’t limited to select highways or specific conditions – it works wherever it’s available It can handle your daily commute or drive you across the country





A Cybertruck just beat a Porsche 911 in a quarter mile while towing another Porsche behind it. The future is electric and it's dragging the past behind it...

If you want to know who has a chance to compete with FSD, apart from the nasty quantization tail that can only be solved with Hardware In the Loop for training, you should also look at the compute. Elon mentioned that you need one H100 to simulate one HD camera at real-time speed. That translates to eight H100s for the complete system. Simulation is necessary for Reinforcement Learning (RL), one of the most potent methods for training a neural network. Tesla has roughly 100,000 H100s available in Cortex. This is enough to simulate 12,500 full eight-camera systems in parallel. This is the best-case scenario, excluding gradient descent and all other overhead. Assuming the average driving speed of simulated scenarios is 50 km/h, you can simulate about 100 million km per week. However, you need to run each scenario several times — once per gradient descent iteration (or epoch). So if you assume 100 epochs (a reasonable ballpark guesstimate), you are down to 1 million km of RL training per week. If you want to reach superhuman safety, you need to train over several million kilometers. Considering you will need many iterations to add new scenarios, improve the reward function, try different network architectures, etc., you cannot get this done within a reasonable time if you do not have this amount of compute. Only Tesla has it. Nobody else is close. PS: To avoid any misunderstanding, I want to stress that I do not mean that millions of kilometers of data capture are enough to develop FSD. You need billions of miles captured by your fleet to discover the critical scenarios needed to train FSD (I explained that in a previous article).






