Scaling Laws

When LLM populations agree, it looks like collective intelligence. A new paper shows it can be amplified sampling noise — a lottery, not reasoning.