Fire-and-forget AI tools have to be much more rigorous than human-in-the-loop tools.
- Fire-and-forget AI tools have to be much more rigorous than human-in-the-loop tools.
- When the human is in the loop, they can catch small mistakes before they compound out of control.
- A human on the loop system must be default-converging.
- That requires approaches like having agents spar and critique one another's work, GAN-style approaches, asking the five hows, etc.