Ex-xAI researcher Ethan He: finding the right axis to scale matters more than raw compute
ricklamers · x · 2026-09-07
Ethan He, formerly of xAI, notes that "just scale it up" is easier said than done — the crucial part is finding the right axis to scale. Large compute clusters existed long before deep learning, but compute went to scientific computing, weather forecasting, and simulations, and GPUs prioritized FP64. Deep learning is what made AI clusters scale to unprecedented levels.
More from AGI Musings
- MIT Tech Review: as AI makes output cheap, judgment becomes the real edge — Olivier__OG · 2026-09-07
- Gary Marcus amplifies theory: delayed Anthropic IPO means racing cheap open-source models — GaryMarcus · 2026-09-07
- "Only 3 years": AI progress outpacing public perception, most still think GPT-3.5 is the norm — RachelVT42 · 2026-09-07
- Math community will pivot to pro-AI within a year, claims researcher Rex Douglass — RexDouglass · 2026-09-07
- The case against conversational interfaces: natural language is not the next paradigm — julianlehr · 2026-09-07
- Vercel CEO Guillermo Rauch: His Agent Runs Product Reviews, He Only Joins When It Fails — cen6wkf · 2026-09-07