Self-driving cars aren't nearly a solved problem (strangecosmos.substack.com)

In a revealing interview, Andrej Karpathy, former head of Tesla’s Full Self-Driving software, emphasized that self-driving technologies are far from being fully autonomous. Despite making substantial progress since the 1980s, the gap between successful demos and practical implementation remains vast, primarily due to the high stakes involved. Karpathy noted that achieving high reliability in self-driving systems is a “march of nines,” where each incremental improvement requires significant effort. He highlighted that many existing deployments still rely on human oversight, indicating that true autonomy is still a long way off. The struggles of major players like Waymo and Cruise Automation underscore the complexities of making self-driving cars a reality. Despite initial optimism and heavy investment, companies continue to face operational limitations. For instance, Waymo operates a modest fleet of around 2,000 vehicles and remains cautious about scaling, which raises questions about the technology's economic viability. The trend of companies abandoning self-driving initiatives, pivoting to other areas, and seeking external funding further illustrates that the path toward fully autonomous vehicles is fraught with challenges, emphasizing the need for ongoing research and development in the AI/ML community.