RE “Agents” Yes, this is the strategy we discussed in the mid 00s and the devs f

RE “Agents”
Yes, this is the strategy we discussed in the mid 00s and the devs from MSFT wanted me to take to DARPA. IMO the model we were discussing captured causality and LLMs don’t but can evolve it. But as I’ve said, LLMs will gradually build causality from top down rather than (as in biologically) bottom up, and they don’t preserve context, where the agents BUILD contexts. If the method using agents can save the resulting manifolds, that begins to create episodic memory FROM work in the LLM universe. At that point we begin to ‘get there’.

cc: @BrianRoemmele