二十天后：Gemini 回答中品牌提及历史预测的预注册测试
原标题：Twenty Days Later: A Pre-Registered Test of History-Based Forecasts of Brand Mentions in Gemini Answers
AI 摘要
一项预注册复现实验测试了品牌在 Gemini 回答中被提及的预测持久性。研究向 gemini-3.1-pro-preview 发送相同的 162 个请求（54 个问题，覆盖英语、爱沙尼亚语和芬兰语），在 2026 年 9 月 22-23 日约十小时内分三波收集，并在 20.3 天后的 10 月 13 日再次收集。基于前九次回答的预测在 Brier 分数 0.0281 和 log loss 0.126 上表现良好，优于品牌历史、总体率和常数 0.5 基线，表明品牌提及率在约 20 天内保持稳定。
正文节选
Twenty Days Later: A Pre-Registered Test of History-Based Forecasts of Brand Mentions in Gemini Answers Abstract Counts of which brands a large language model names are useful only if they hold long enough to act on. We test this with a pre-registered replication. The same 162 requests (54 questions in English, Estonian and Finnish about three software categories, four of twelve candidate brands per question, three repeats) were sent to gemini-3.1-pro-preview in three waves within about ten hour