X @Herbert Ong

Autonomous Driving Technology & Safety - Adding more diverse sensors does not necessarily equate to increased safety, and could potentially decrease safety due to conflicting data sources and decision-making complexities [1] - Aviation utilizes sensor redundancy with triplicate sensors for critical systems, allowing for easy isolation of faulty sensors, a method not followed by Waymo [1] - Waymo's reliance on diverse sensors, algorithms, and high-definition maps introduces latency and potential errors in data processing, which could lead to crashes [1] Tesla's FSD & Approach - Tesla's Full Self-Driving (FSD) operates as a RoboTaxi in Austin using vision and end-to-end neural nets, enabling easier learning from video and auto-labeling compared to LiDAR point clouds [2] - Tesla's vision-only neural net FSD is considered a more elegant and simpler solution, aligning with the engineering principle of "less is more" [4] - Tesla's FSD benefits from a long context window and occupancy network, allowing it to remember occluded objects, enhancing its perception capabilities [3] - Tesla's FSD will improve with increased compute power from AI5 hardware (mass production in Q1 2026), and more training runs on its model using Tesla's 50K Nvidia's and 20K Dojo Cortex AI Compute center [4] Competition & Performance - Tesla's FSD Supervised operates in 5 regions (US, Mexico, Canada, China, and Puerto Rico) with pending regulatory approval for more, while Waymo operates in only 4 US cities [3] - Tesla's end-to-end neural net solution is more scalable and adaptable compared to Waymo's approach, which involves multiple sensor types, high-definition maps, and rules-based code [5]