Zeeshan Pishehvar: Understanding Superintelligence Scaling Requires Viewing Training as One Compute Flow

ai · x · 2026-11-23

Quoting Elon Musk's quip that pretraining should just be called 'training', Zeeshan Pishehvar argues something deeper is at stake: understanding the scaling properties of superintelligence requires viewing the entire training pipeline—from pretraining to RL—as one singular flow of compute and data. Musk also predicts data starvation will be next year's core training challenge.

Related event: Debate with Musk Over 'Pretraining' Naming and Looming Data Exhaustion(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →