AI scaling laws

Relationships describing how model performance scales with increased parameters, data, and compute. Typically follow power-law patterns (e.g., loss ∝ N^-α for parameter count N). Key implications:

• Predictable performance gains from scaling
• Efficiency trade-offs between data/compute/parameters
• Basis for model development strategies

Recent Discussions

• Mixture of Experts podcast panel (hosted by Tim Hwang) featuring:
  • gabe-goodhart (Chief Architect, AI Open Innovation)
  • Abraham Daniels (Sr. Technical Product Manager, Granite)
  • Aaron Baughman (IBM Fellow, Master Inventor)
• Key debate points:
  • Validity of traditional scaling laws amid “Fun-cember” model releases (end-of-year surge in major model launches)
  • Analysis of amazon blocking chatgpt as indicator of competitive dynamics
  • Questions about diminishing returns in scaling efficiency

2026 04 14 Ibm panel

Source Notes

• 2026-04-23: Engine Survival: The Critical Role of Oil Pressure and Warning Lights
• 2026-04-14: [[lab-notes/2026-04-14-Optimizing-AI-Costs-and-Privacy-with-Local-Open-Source-Models-and-Hybr|“But OpenClaw is expensive…“]]