Xpeng AI chief: language is poison for autonomous driving; $41M/month AI training, world model integration; Tesla tests steering-wheel-free Cybercab on public roads
Key Questions
What is Xpeng's stance on using language models for autonomous driving?
Xpeng's AI chief states that language as an intermediate output is harmful, with the company spending $41M monthly on AI training and planning world model integration later this year.
What testing has Tesla conducted with the Cybercab?
Tesla is testing production Cybercab vehicles without steering wheels on public roads in Austin, with a safety monitor present; minor collisions have been reported amid ongoing NHTSA discussions.
How does Tesla's FSD safety performance compare in recent data?
Recent stats indicate one major crash per 5.3 million miles with FSD, providing context on performance alongside camera-only approaches versus competitors using lidar and radar.
Xpeng's AI chief argues that using language as an intermediate output is harmful for autonomous driving. The company spends $41M per month on AI training, claims parity with Tesla FSD v13, and plans to integrate a world model later this year. Radar is retained only for safety redundancy. This is a key signal for autonomous driving progress and competitive dynamics. New: Tesla tests production Cybercab without steering wheel on Austin public roads – safety monitor present, NHTSA proposal to allow, minor collisions reported. Camera-only vs Waymo's lidar/radar. Key step for Tesla's autonomy ambitions but still far from mass deployment. Added: FSD safety roundup with stats (one major crash per 5.3M miles) – surface-level but adds context.