Humanizing LLM Outputs Is Dumb
Humanising LLM Outputs Is Dumb
The author argues that viral prompts like 'I have ADHD' or 'use Simplified Technical English' are the wrong way to fix LLM verbosity. Compressing output during generation is lossy, hides failures, and breaks agent-to-agent communication. Instead, agents should keep high-fidelity state and only compress at the human boundary, turning these viral hacks into a bug report for better tooling.
The durable version is agents whose native language is precise, machine-facing state, with the warm, concise, human version generated only at the boundary.