“According to Jeffrey Irving, Chief Scientist at the UK AI Security Institute, current frontier AI models can troubleshoot laboratory experiments from a cell phone picture more effectively on average than human PhDs.”