For robots, a drop in reliability is itself an accident

This is a write-up of a 48-minute recording that follows automation's move into the physical world through four areas: semiconductors, compute facilities, humanoids and factories. The sharpest distinction is tolerance for error: large language models came out into the world even at 60–70% reliability, and hallucinations did little harm, but for a moving robot, it is stated flatly, a drop in reliability is itself an accident. It then presents field survey findings that 70–80% of on-site work cannot be done with a gripper and requires a human-level five-fingered hand and dexterity, adding the observation that the robots shown dancing and fighting have no hands. The yardstick for reading specs is turned upside down as well: moving fast is actually easy, while standing still or holding something for a long time is hard and generates heat; on the equipment side, a flow is demonstrated in which the operator asks about a shuttle's status in natural language, gets a sensor misdetection back as a candidate cause, and then orders a transfer. However, the export and power figures are all projections without sources, so it is safer to take only the framing and the examples.

This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.