Reality Check: AI photo detection game (realitycheckk.com)

Reality Check is an interactive web game that asks players to identify which photo in a short batch is AI-generated by clicking the image or using the arrow keys. The interface cycles themed batches (the current theme is “messes”), shows an average player accuracy (currently 78%), and refreshes with a new batch on a daily cadence (next batch in ~1d 23h 38m). The experience is intentionally simple — pick the synthetic image from a small set — and aggregates player performance in real time. For the AI/ML community this is a lightweight, gamified probe of human ability to spot synthetic imagery and a potential source of labeled examples for benchmarking detectors or studying human–AI complementarity. Technical takeaways: theme-curated batches can reveal where detection fails (specific scene types or artifacts), a daily refresh enables longitudinal tracking, and aggregated accuracy provides a quick human baseline for comparison with automated detectors. Caveats include selection bias (curated themes, short tasks) and limited context compared with real-world forensics. Still, Reality Check is useful for public education, quick human-baseline estimation, and — if data are collected and shared responsibly — building or evaluating detection models and human-in-the-loop defenses against image synthesis.