Reading Log

Ben Thompson's clearest articulation of why compute access is becoming the defining competitive advantage in AI. The analogy to oil economics is surprisingly apt.

A thoughtful breakdown of why agent architectures are harder than they look. The key insight: reliability at 95% vs 99.9% is the difference between a demo and a product.

Power law dynamics in venture explained with actual data. A good reminder that the job isn't to avoid losses — it's to find the outliers.

Sutton's argument that general methods leveraging computation always win over human-designed features. Written in 2019 but more relevant than ever in the age of large language models.

PG at his best — practical advice on finding and doing meaningful work. The section on the importance of working on problems that feel like play resonated deeply.

Nathan Benaich and Ian Hogarth's annual report remains the best single overview of where AI stands. Essential reading for anyone in the space.