LLMs are the Jungian unconscious plus Freudian repression (RLHF) alternate model from the shoggoth thing. (CWT 218 Paul Bloom)
Rats not understanding prime number mazes, but getting even number mazes, is a good example of hard limits to understanding from intelligence levels.
Are beliefs and preferences stored differently in LLMs?
Does LLMs as simulators raise the status of exemplarist virtue ethics? (Check notes I think I’ve already thought this)
Don’t fall into the trap of anthropomorphizing LLMs and assuming that failures which would discredit a human should discredit the machine in the same way. - SW
Weird old philosophy (of mind)
Re the question of how much compute has gone into training the human brain over the course of evolution vs. within a life. How much compute is in the environment or something:
- Soviet psychologist and prominent “pre-cognitivist”. His work focused on the development of cognitive skills, especially skills of self-control, and how these were supported (“mediated”) by external tools, above all the culturally-provided tools of language and social interaction. He was particularly interested in the ways people “internalize” such tools, learning to do without such scaffolding, though it’s necessary for the acquisition of the skill. (He went so far as to speculate that all cognitive abilities originated as internalizations of social interactions. Whether he meant this ontogenetically or phylogenetically, or in some sense both, I can’t tell, and in any case it seems very unconvincing to me.) One way of thinking about what he was doing (grossly anarchonistic but useful) is that he was interested in how orgnaisms with limited computational ability can effectively expand their information-processing powers by interacting with structured environments --- think of how, in formal language theory, attaching a stack to a finite-state machine lets it generate context-free languages, not just regular languages. Another anachronistic framing is that he was interested in collective cognition. Some modern Vygotskyans believe their socio-cultural approach is an alternative to the more usual, computational approach to cognition. This has never made any sense to me, more or less for the reasons Frawley lays out in his book.