Google Vantage uses AI role-play to assess collaboration under pressure

几千年都没考过这个？谷歌「最毒」AI考局，专测你在压力下怎么做人

Google Research and NYU tested Vantage with 188 US participants aged 18-25 on conflict resolution and project management. Its four-layer agent pipeline generates scenarios, applies pressure, extracts behavior, and scores against rubrics; AI-human agreement matched expert-expert Kappa of 0.45-0.64. The key gap is transfer beyond lab settings; the post says Vantage remains a Google Labs research experiment.

Why it matters: HKR-H/K/R all pass: the Vantage study has a strong hook, concrete sample size, and evaluator-risk resonance. It stays in the low featured band because it is still a Google Labs experiment with a narrow cohort.