Open weights need orders-of-magnitude cheaper training or it's a LARP, researcher argues

jd_pressman · x · 2026-09-13

jdpressman argues that for open weights to survive, model training and inference must get several orders of magnitude cheaper — anything else is "a LARP." He clarifies this isn't wishful thinking: he finds the convergence-to-giant-brain scenario dystopian (suggesting e.g. the end of sex), and doubts superintelligence will ever run on a 2016 MacBook, if ever.

Original post →

Open weights need orders-of-magnitude cheaper training or it's a LARP, researcher argues

More from AGI Musings