The people who trained Tesla’s self-driving AI won’t ride in it
Workers Who Trained Tesla’s Self-Driving Software Share Their Concerns
A Reuters investigation revealed a concerning trend among the workers who trained Tesla’s self-driving software. Out of nine former Tesla data labelers and a former self-driving engineer interviewed, seven stated they would not ride in Tesla’s Full Self-Driving (FSD) mode.
Key Findings:
- Routine Speed Violations: Five data specialists confirmed seeing clips of Teslas exceeding speed limits while operating on FSD.
- Low Priority for Speed Issues: Engineers and managers prioritized edge cases over common speeding problems, demonstrating a potential gap in the system’s safety measures.
- Internal Footage Concerns: The data labelers’ testimony suggests that internal footage may contain even more failures than those reported in media.
Background:
Tesla has been expanding FSD availability globally, including China, though the unsupervised version remains in trial phase only on robotaxis in Austin, Texas. Recent FSD-related incidents, such as vehicles driving into lakes and bridges, have raised safety concerns.
The data labelers’ unfiltered perspective highlights a critical issue: the gap between Elon Musk’s claims of fully autonomous driving and the actual performance of the system. Waymo’s recent shutdowns due to flooding further underscores the complexities of developing robust autonomous driving technology.