Post

Seriously: is the next step in evolution AI principals? AI principals would be agents that not only execute tasks but also define strategies, goals, and policies—essentially, they would take on a higher-order role.

In my opinion, those "higher order" roles for agents today require way too much context that LLM model driven decision making just won't every work well for. We need another once in a generation breakthrough to make it that good.

0 reposts1 like

Yeah I think you might be correct. At least for currently publicly available models, the deterioration on context is pretty obvious. I actually think the average lifespan of my LLM sessions is going down, not up.