SealedBench: A Protocol for Freezing AI Benchmarks Before the Questions Are Known

This story was originally published on HackerNoon at: https://hackernoon.com/sealedbench-a-protocol-for-freezing-ai-benchmarks-before-the-questions-are-known.
An AI exam whose generator is sealed to a future drand round and stored on Arweave, so anyone can rebuild the questions and re-check the scores.
Check more stories related to web3 at: https://hackernoon.com/c/web3. You can also check exclusive content about #decentralize-ai, #decentralize-ai-hackathon, #arweave, #decentralized-storage, #verifiable-ai, #llm-evaluation, #sealedbench, #ai-benchmarks, and more.

This story was written by: @ZGZ_rrdpeg7. Learn more about this writer by checking @ZGZ_rrdpeg7's about page, and for more stories, please visit hackernoon.com.

I sealed an AI exam generator with drand timelock encryption and stored it on Arweave, with the exact models, prompts and scoring rules frozen beforehand. When the beacon round arrived, it unlocked the generator and seeded the questions, and four open-weight models answered them on a CPU. On fresh questions Falcon3-1B scored 47/80, against 21/80 on public GSM8K. One transaction ID lets anyone rebuild the questions and re-score the answers.