Study design, program evaluation, statistical analysis, AI evaluation, and working with statisticians.
This week's AI-evaluation research asked what a score really measures: cheaper accuracy audits, coding agents judged against a narrow reference, fact-checking scores that hide gaps, and a ranking metric that masked a failing defense.
In 1930, Scotland gave milk to 10,000 schoolchildren and compared them with 10,000 more. Kind teachers had already tilted the result.
A public depression-trial dataset shows how independent calculations, follow-up uncertainty, and missing-data assumptions fit together.
This week's AI-evaluation research was about trusting the grader and the score: a self-debiasing LLM judge, screening judges with no ground truth, noisy agent benchmarks, and a vendor-graded search leaderboard.
A 1985 study called the hot hand an illusion. In 2018, two economists found the test itself was biased against streaks.
A clean run proves AI-written analysis code is valid, not that it estimates what you meant. Four checks that produce real validation evidence.
{{ p.excerpt }}
{{ noResultsMsg }}
One email whenever we publish something new. No spam, unsubscribe anytime.
{{ subscribeErrorMsg }}