I basically don’t buy 1 as an agency skill specifically, and I think that a lot of the agents like AutoGPT or Langchain also don’t work for the same reasons that current AIs are not too useful, and I think that improving reliability to how an LLM does it’s work would both benefit the world model that isn’t agentic, and the agentic AI with a world model.
Agree that 2 is more of an agency skill, so it is a bit of a bad example in that way.
I agree your way of solving the problem is one potential way to solve the reliability problem, but I suspect there are other paths which rely less on making the system more agentic.
Re 1: I guess I’d say there are different ways to be reliable; one way is simply being better at not making mistakes in the first place, another way is being better at noticing and correcting them before anything is locked in / before it’s too late to correct. I think that LLMs are already probably around human-level at the first method of being reliable, but they seem to be subhuman at the second method. And I think the second method is really important to how humans achieve high reliability in practice. Hence why LLMs are generally less reliable than humans. But notice how o1 is already pretty good at correcting its mistakes, at least in the domain of math reasoning, compared to earlier models… and correspondingly, o1 is way better at math.
I basically don’t buy 1 as an agency skill specifically, and I think that a lot of the agents like AutoGPT or Langchain also don’t work for the same reasons that current AIs are not too useful, and I think that improving reliability to how an LLM does it’s work would both benefit the world model that isn’t agentic, and the agentic AI with a world model.
I’m more informed by these posts specifically:
https://www.lesswrong.com/posts/YiRsCfkJ2ERGpRpen/leogao-s-shortform#f5WAxD3WfjQgefeZz
https://www.lesswrong.com/posts/YiRsCfkJ2ERGpRpen/leogao-s-shortform#YxLCWZ9ZfhPdjojnv
Agree that 2 is more of an agency skill, so it is a bit of a bad example in that way.
I agree your way of solving the problem is one potential way to solve the reliability problem, but I suspect there are other paths which rely less on making the system more agentic.
Re 1: I guess I’d say there are different ways to be reliable; one way is simply being better at not making mistakes in the first place, another way is being better at noticing and correcting them before anything is locked in / before it’s too late to correct. I think that LLMs are already probably around human-level at the first method of being reliable, but they seem to be subhuman at the second method. And I think the second method is really important to how humans achieve high reliability in practice. Hence why LLMs are generally less reliable than humans. But notice how o1 is already pretty good at correcting its mistakes, at least in the domain of math reasoning, compared to earlier models… and correspondingly, o1 is way better at math.