Videos

Self-driving in general has been overhyped by grifter tech bros like Elon and really shows the current limits of ML. Today, ML models are basically fuzzy, probabilistic functions that map inputs to outputs and are not capable of actual reasoning. There is a long tail of scenarios where a self-driving car will not generalize properly (i.e., will kill people). Throwing increasingly more data and compute at it won’t suddenly make it capable of reasoning like a human. Like other ML use cases, self-driving is a cool concept that can be put to good use under the right conditions, and can even operate mostly without human supervision. However, anyone claiming it’s safe to let today’s “self-driving” cars shuttle humans around at high speeds with no additional safeguards in place either has an unrealistic understanding of the tech or is a sociopath.