[vincnetas]

continuing on personal asistant analogy, i bet even now we have ultra rich people who are not smart enough to do things themselves but are smart enough to buy (hire) smart people to work for them. And even allow them to make decisions without understanding them. But with only a guard rails : does this produce wealth for me. If yes, do what you need, i don't care :)

So this i think is applicable to AI also, pay for smarter than you AI's pit them against each other, let them supervise each other and measure the outcomes you need. Who cares how they achieve that (sound clinical and scary)

[oerdier]

It has gone well enough for these rich people because their hired assistants were human. With humans there is some degree of responsibility, liability and a slow tempo (= time to correct) involved. LLMs are a significantly different beast.