tension compass
Do current LLMs possess a genuine world model / understanding, or only statistical regularities over text form?
T01 genuine-disagreement
LLMs already have (at least graded/limited) world models, meaning, or understanding.
LLMs track statistical form and lack a genuine (grounded/causal) world model; fluency invites overestimation.
POLE APOLE BHofstadterLeivada et al.Safron et al.Yiu et al.Krakauer et al.BreithauptKu et al.Collins et al.
click a position for its verbatim quote