Loading slide…
You will review 30 cases in two parts. The first 10 cases each ask five short multiple-choice questions about statements taken from AI-generated reports, with the reference diagnosis shown. The remaining 20 cases show anonymised AI reports to score — one report at a time, one question at a time, judged against the slide alone.
Your answers are saved automatically after every click. You can close the browser at any time and continue later by entering the same access code.
You have finished all 10 verification cases. We are wholeheartedly grateful. Your verdicts are what lets us measure how far an AI evaluator can be trusted against real expert eyes.
Part 2 is a little more taxing, but it answers the central question of this study:
Which vision–language model writes the best and the worst
pathology reports?
PRISM · PRISM2 · SlideChat · MI-Gen · HistoGPT · SlideFlame
You will score the reports for 20 cases. The reports are anonymised and shuffled. You will never know which model wrote which report, and the reference diagnosis stays folded away: the slide alone is the judge.
Take a break first if you like — your progress is saved.
Thank you! Your answers have been saved.
Please also press the button below and send the downloaded file back to us, as a backup.