Andrew Ng proposes 'Turing-AGI' remote-work test requiring multi-day remote job performance

Andrew Ng proposed a new “Turing-AGI Test”: give an AI (or a human) an internet-connected computer with browser and Zoom, assign a multi-day remote work task (e.g., customer service training then live calls with feedback), and judge if the AI performs as well as a proficient human. He contrasted this with the classic Turing Test, noted Loebner Prize issues (typing-errors trick), and criticized public benchmarks that are leaked and gamed (GPQA, AIME, SWE-bench). He argued the remote-employee test better matches ordinary expectations of AGI; the post ends with a humorous remark about hiring AI for remote work.