A benchmark for humans: rank people by turns and tokens burned prompting an LLM
DominiqueCAPaul · x · 2026-09-16
A proposal for a "human benchmark": have people prompt an LLM and rank them by (a) how many human-LLM turns it takes to reach a solution and (b) how many tokens they burn. Essentially turning prompt efficiency into a quantifiable competition.
More from coding & agent
- A long-read mapping the agentic AI ecosystem in ten layers, from models to knowledge — davletdz · 2026-09-16
- An agent read my planning notes as approval and shipped 31 unsupervised changes — AnvilandCode · 2026-09-16
- Supabase Select 26 lineup: Steve Wozniak, Garry Tan, and Anthropic execs join Oct 2 event — dshukertjr · 2026-09-16
- Capawesome Ships Three MCP Servers to Stop AI Assistants Suggesting Outdated APIs — DayanaJabif · 2026-09-16
- Synthesia launches real-time Interactive Avatar API on LiveKit — synthesiaIO · 2026-09-16
- Founder ditches $10/mo Super.so as AI agent migrates 100-page site in 5 minutes — thisiskp_ · 2026-09-16