MasonKamb: gradient descent is the 'original sin' behind LLM-human cognition divergences

_arohan_ · x · 2026-09-06

MasonKamb shares a self-described crackpot opinion: gradient descent is the "original sin" behind many divergences between LLM and human cognition. arohan quotes and agrees, arguing there would be a lot more progress if we rethink gradient descent for training.

Original post →

More from AGI Musings

AGI Musings channel →