Modern AI Training Moves Beyond Pure Prediction, Mooting Transformer Debate
——0——
Modern AIs are trained for a while on pure prediction, and then they're trained on solving hard problems. Roughly speaking, someone will give them a math problem and 1000 tries to write a bunch of text about solving it, and then pick the closest attempt & reinforce it.
Nate Soares ⏹️@So8resBut the debate about whether a transformer architecture trained on pure prediction with reasonable amounts of compute winds up superintelligent is moot. Because modern AIs aren't trained on pure prediction anymore.
7:35 PM · May 21, 2026 · 140 Views
7:35 PM · May 21, 2026 · 145 Views