Why LLMs can't count
Are they System 2 processors yet?
Topic
How AI Actually Works
You cannot steer a technology you do not understand. These pieces open up the technical workings of modern AI: its architectures, its interpretability, and where today’s models break.
Why is AI shale-shaped?
The material contingency of the AI boom
The Post-Transformer Era
3 architectures competing to replace the transformer
Why AI can't learn on the job
And how that might change
Hallucination solutions, China's AI strategy and AI companions
Is AI overhyped, and the challenge of compute-based thresholds, TFP Newsletter #2
The AI boring law
Why LLMs don't have 'negative capability'
Why we don't understand AI models yet
Breaking apart LLMs with Lee Sharkey