tweet-scaling-axes-history

Four Scaling Axes: History

Source: https://x.com (tweet) Author: unknown Date: 2025-2026

Summary

A genealogy of four AI scaling axes attributed to specific individuals: Alec Radford (1st: pretraining compute), John Schulman (2nd: RLHF/alignment), Noam Brown (3rd: test-time compute/search), and Peter Steinberger (4th: agentic/real-world tasks).

Key Claims

• Alec Radford: 1st scaling axis — pretraining data/compute (GPT-1,2,3)
• John Schulman: 2nd scaling axis — RLHF, scaling human feedback signal (ChatGPT)
• Noam Brown: 3rd scaling axis — test-time compute (o1-style reasoning)
• Peter Steinberger: 4th scaling axis — agentic/interactive task performance (Cursor?)
• Each axis represents a new paradigm shift in what can be scaled

Concepts

• Scaling & Compute — the four axes of AI scaling history
• Test-Time Compute — Noam Brown’s contribution (3rd axis)
• RL Infrastructure — Schulman’s RLHF as 2nd axis