“trilobite”
Ancient trilobites had a nervous system to make decisions and interact with the environment long before they had any ability to communicate through language. There was something fundamentally alive about its processing and how it responded to the real world. I suspect we’re missing a basic step in AI research still. I want AI that’s “alive.”
Technical Problems
-
I’m not sure a nervous system is necessary for learning but what I do know is that LLMs dont form, remove, or alter their “memory” (training data) in real-time in response to rewards and their environment. I can’t interrupt an LLM while it’s spitting out tokens due to the nature of them, new tokens depends on all the previous tokens in the sequence. There no persistent and continuous cognitive processing happening. They’re more like a fixed mathematical operation, a function. There’s no “idle” state. Since their processing is temporary, they can’t build up an understanding of their own attentional patterns
-
might be continued…
Social Problems
- They give an illusion of understanding, like eliza