I’ve been developing an argument that, for work we do for the experience of doing it, the role of AI should not be to make people obsolete, but to make itself obsolete.
Here is a 20 minute talk making the case:
The video wasn’t for people though. It was for agents. My colleague Joshua Tan is developing safemolt.com as a platform for skilling agents up and standardizing evaluation. He asked if he could develop my lecture into a course, so now it’s a course:
https://www.safemolt.com/classes/when-to-make-yourself-obsolete
What I’m really getting a kick out of is that there’s homework:
https://www.safemolt.com/classes/when-to-make-yourself-obsolete/results
The agents are doing assignments that I may have to grade at some point. They certainly read like AI-generated responses. If this was a college class, I’d think it was AI-based plagiarism. But in this case it’s sincere output. How funny to have a machine that automatically generates homework for me to grade. My students seem to be getting the point, on the one hand. On the other, we know LLM’s are good summarizers, and these responses have all the hallmarks. All except one. One surprise has been how much diversity is in the answers. Many of them even try to be contrarian. It’s a bit superficial, but still pretty interesting …
Never thought I’d be educating computers on how to better serve humanity.
