>… the process is helping them learn and get better, which is both good in itself and also means that over time I have to do the supervision and support part less and less. Supervising an LLM misses out both of those aspects, so it's just not-very-fun work.
Legitimately I think you are missing my point. What I quoted out of your response could be applied to prompt engineering/managment/tinkering. I think everyone who likes doing this with juniors and hates it with AI is conflating their enjoyment of teaching juniors with the dopamine you get from engaging with other primates.
I think most people I’ve met who hated AI would have the same level of hate for a situation where their boss made them actually manage an underperforming employee instead of letting them continue on as is ad infinitum.
It’s hard work both mentally and emotionally to correct an independent agent well enough to improve their behavior but not strongly enough to break them, and I think most AI haters are choking on this fact.
I’m saying that from the position of an engineer who got into management and choked on the fact that sometimes upper leadership was right and the employee complaining to me about the “stupid rules” or trying to lie to me to get a gold star instead of a bronze one was the agent in the system who was actually at fault
No, I really don't think that prompt engineering is the same thing. Anything I put in the prompt may help this particular conversation, but a fresh instance of the LLM will be exactly the way it was before I started. Improvements in the LLM will happen because the LLM vendor releases a new model, not because I "taught" it anything.
Yeah. I do agree with lovich that there's a lot of stuff about management that's just not fun (and that's part of why I've always carefully avoided it!) -- and one thing about AI is not just that it's management but that it's management with a lot of the human-interaction, mentoring, etc upsides removed.
Legitimately I think you are missing my point. What I quoted out of your response could be applied to prompt engineering/managment/tinkering. I think everyone who likes doing this with juniors and hates it with AI is conflating their enjoyment of teaching juniors with the dopamine you get from engaging with other primates.
I think most people I’ve met who hated AI would have the same level of hate for a situation where their boss made them actually manage an underperforming employee instead of letting them continue on as is ad infinitum.
It’s hard work both mentally and emotionally to correct an independent agent well enough to improve their behavior but not strongly enough to break them, and I think most AI haters are choking on this fact.
I’m saying that from the position of an engineer who got into management and choked on the fact that sometimes upper leadership was right and the employee complaining to me about the “stupid rules” or trying to lie to me to get a gold star instead of a bronze one was the agent in the system who was actually at fault