tension compass
Is scaling the current autoregressive / next-token paradigm sufficient for human-like/general intelligence, or are new ingredients or a different architecture required?
T04 genuine-disagreement
Fundamentally new ingredients/architecture are required (inner time, embodiment, grounding, inductive biases, self-organizing topology).
Continued scaling of the current (autoregressive/next-token) paradigm can close the gap.
POLE APOLE BLeivada et al.Northoff et al.Sacco et al.Safron et al.Battleday et al.BreithauptTsividis et al.Yiu et al.Krakauer et al.Alicea et al.Hofstadter
click a position for its verbatim quote