This does not in any way prove that LLMs aren’t sentient. Feels like a joke paper. I mean not a completely unserious one but one that’s has been written with a lot of smirk.
The actual author explains it way better though: https://adriandewynter.substack.com/p/if-llms-have-human-like-attributes
I obviously didn’t understand everything, but it seems to me that he built an LLM in AoE2 goats and basically said “look, obviously this isn’t concious” and that “obviously” is doing much more work than it should. Reductio ad absurdum.
Obviously the LLM built in AoE2 goats is conscious.
Look, if the AoE goats are as evil as real goats then the real danger to humanity has been created here, not by the big AI players. Fear for all we hold dear, comrades.
I made a cat-powered MLEM in a game that has cats in it to prove that those furry bois are pettable
Wtf does goats mean here?
Simulated goats in the game Age of Empires II.
Ok having never played this one, how is the AI in this case powered by the in game goats?
Check out the circuits here: https://adewynter.github.io/notes/aoe2-circuits
Apparently, he builds a functioning (probably a very basic) LLM from those.
Yea I was just reading through that. Thanks for the direct link. He conveniently leaves out any details of the size of the “LLM” he “trains”. Seems like a cool science experiment but a huge leap to extrapolate the premise from.
