Hinge #2 bet proposed in this post: https://www.lesswrong.com/posts/45oxYwysFiqwfKCcN/untitled-draft-keg3

"If the fastest models trained in mid‑2027 require one‑quarter the elapsed time of equally large runs eighteen months earlier without a node shrink below four nanometres, that will be strong evidence that software‑only acceleration is real, and is not highly bottlenecked by compute or real-world experiments slowing the AI researchers down."

The other bets:

#1: https://manifold.markets/rayman2000/nvidias-datacenter-revenue-and-bigt

#3: https://manifold.markets/rayman2000/revenue-per-deployed-h100-exceeds-1

#4: https://manifold.markets/rayman2000/g7-country-manages-three-years-of-6

#5: https://manifold.markets/rayman2000/a-topthree-ai-lab-delays-a-frontier

Meta-market for majority of the five bets: https://manifold.markets/rayman2000/will-a-majority-of-the-5-hingequest?play=true