Success Per Tokens

According to LessWrong, the Pareto frontier — a concept from optimization theory — is reshaping how we evaluate AI models and, by extension, human productivity and business strategy. In AI, the question is no longer just "how capable is this model" but "how capable per token spent." Frontier language models show logarithmic performance curves: early compute yields big gains, but eventually performance plateaus regardless of budget. Some models optimize for cost by reducing token output in reasoning modes, trading performance for affordability. This framework applies elsewhere too. People can improve their efficiency with AI by learning tools, automating workflows, and writing better prompts instead of relying on brute-force model calls. Companies face similar tradeoffs: startups often outpace incumbents by optimizing for growth with minimal capital, especially when their product is digital. But as companies scale, management complexity and competitive pressure eventually slow that growth. The recurring pattern: across AI, individual work, and business, efficiency per unit of resource invested often matters as much as absolute capability.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton