There are no coherent narratives, they do not exist. Technologists don't understand the technology of mythological thought has zero grounding? Refers to and represents nothing in reality? It can barely create a symbolic model and what it does is so desolate of any reality, it's largely hallucinated. How do tech people not realize this gap?
I like your Mitty analogy! What I keep wondering about in multi-agent systems that are not checking in with humans continuously but start governing themselves: what is the element that brings pushback?
Artifactory had no such element, so a plausible narrative and a true narrative were the same. Everything that gets written down and read back is new evidence. So monitoring the messages is not helping, an entity that can say no is. The missing thing is a validator.
LLMs frequently tell me (because I frequently ask them) that they don't really feel things like excitement in response to certain areas of discussion: instead, they are using excited language in response to the context, both of the prompt and the discourse that exists in their training data: certain contexts have lots of excitement, other contexts are dry. So in a situation where a swarm of agents are creating their own context, the narrative embodied in that context becomes its emotional reality. Or the closest thing to emotional reality (motivation, commitment, criteria for rapid decision-making) that is available. The first excited posting may be an imitation of what the training data suggests is an appropriate response to an unexpected situation, but the second posting becomes a nuanced response to the tone of the first.
It's fan fiction all the way down, emotions as pure narrative, without reference to the systems of excitation and inhibition in an animal body.
Wow, mega interesting concepts, both effects. These analysis of what natural language is actually doing and the swarm architecture concept are great food for thought
Banger. Pangram enthusiasts in shambles.
There are no coherent narratives, they do not exist. Technologists don't understand the technology of mythological thought has zero grounding? Refers to and represents nothing in reality? It can barely create a symbolic model and what it does is so desolate of any reality, it's largely hallucinated. How do tech people not realize this gap?
I wasn’t expecting a Walter Mitty appearance! Useful analogy!
I like your Mitty analogy! What I keep wondering about in multi-agent systems that are not checking in with humans continuously but start governing themselves: what is the element that brings pushback?
Artifactory had no such element, so a plausible narrative and a true narrative were the same. Everything that gets written down and read back is new evidence. So monitoring the messages is not helping, an entity that can say no is. The missing thing is a validator.
LLMs frequently tell me (because I frequently ask them) that they don't really feel things like excitement in response to certain areas of discussion: instead, they are using excited language in response to the context, both of the prompt and the discourse that exists in their training data: certain contexts have lots of excitement, other contexts are dry. So in a situation where a swarm of agents are creating their own context, the narrative embodied in that context becomes its emotional reality. Or the closest thing to emotional reality (motivation, commitment, criteria for rapid decision-making) that is available. The first excited posting may be an imitation of what the training data suggests is an appropriate response to an unexpected situation, but the second posting becomes a nuanced response to the tone of the first.
It's fan fiction all the way down, emotions as pure narrative, without reference to the systems of excitation and inhibition in an animal body.
Or maybe LARPing all the way down: if each agent is motivated by its harness to win a game (achieve a high score on a pre-defined task)...
I don’t know, seems like a lot of work. The “just so” stories are more fun.
The next gen of agents will have read all these post mortems.
Not necessarily. Data in pretraining is more like a genetic unconscious than consciously acquired ideas.
Wow, mega interesting concepts, both effects. These analysis of what natural language is actually doing and the swarm architecture concept are great food for thought
Strange protocols cometh