With Cheaper Reasoning and Mass Copying, ASI May Arrive Before AGI

MickeySteamboat · x · 2026-07-22

The author argues that by taking a model capable of long reasoning, creating thousands of its copies, and dramatically reducing the cost of that reasoning, we can point this collective power at major human problems like math, new medicines, energy, and materials.

Given the capability jumps seen in current models and the untapped order of magnitude in hardware efficiency, the author predicts that before the end of this year, we might see a form of uneven Artificial Superintelligence (ASI) before achieving Artificial General Intelligence (AGI). This intelligence would be strangely limited in some areas while far beyond human capabilities in others.

Related event: Extended Inference and Plunging Costs Could Accelerate ASI(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →