How the AI Report Card is made

Every day we gather the previous 24 hours of news about AI, compute, energy, nuclear, fusion and quantum from newsrooms, filings, labs and researchers worldwide. Each story is verified against its own sources before it is rated.

One question

Each story is rated from −3 to +3 on a single question: does it make AI safer, more useful, more trustworthy and more widely available for people everywhere? The day's grade is the weighted average, where bigger and more widely reported stories count more.

Checks

Claude writes and rates the report. An independent agent re-opens every source. A council of ten AI models, the flagship of each of ten leading companies in the US and China, re-rates every story without seeing ours, flags facts it doubts, and can send a rating back for a second look.

Five ways to read it

The executive brief, general, hobbyist, professional and expert views show the same facts and ratings. They differ only in depth and wording. General-reader wording is written to add nothing the original summary doesn't say.

Sponsors

Sponsorship slots are sold the evening before in a sealed auction, without sight of the report. Sponsors have no say in any story or rating, and when a report covers a sponsor, readers are told beside the sponsorship. How sponsorship works