Featured in Domain Arrivals · 2026-08-08
Funding Bench Asks Whether LLMs Can Judge Grant Proposals
Funding Bench tests language models on historical AI-safety grant proposals, asking them to recommend funding and predict each project's future fundraising.
Why it surfacedThis is a concrete benchmark rather than another vague AI demo: five models assess 20 proposals across four funder rubrics, with outcomes hidden and forecasts scored against subsequent funding through 2028. It turns a live question about machine judgment into a small, inspectable experiment—and flags famous projects whose outcomes models may have memorized.
Continue into the edition
The Fax Is Thinking
August 8’s newborn domains include a chatbot that replies only by fax, an Elmira tax analysis built from 1,689 sales, a map protecting Jeju divers’ precise fishing grounds, and 58 boxes catalogued into one uncle’s 1,113-book library.
This is one of 1,000 discoveries in the 2026-08-08 Domain Arrivals issue.