linkdrift
Pretraining and scaling: a universal methodology

Pretraining and scaling: a universal methodology

Jiaxuan Zou
@SmartPig_Joe·1.9K followers
·10 min read

Originally published on my personal blog on September 7, 2026, and revised on September 8, 2026. This version includes edits for readability and formatting, plus dated additions made on November 3,

Originally published on my personal blog on September 7, 2026, and revised on September 8, 2026. This version includes edits for readability and formatting, plus dated additions made on November 3, 2026. Read the original article. Three takeaways 1. Viewing pretraining and scaling as a universal methodology, we can predict that test-time scaling should emerge in embodied intelligence too. Train-time scaling is already taking shape there. 2. LLMs are criticized for being too slow for embodied tas

197
0
2
0
Read full article on X