elvis· @omarsar0 · X·· 2 小时前AI 评分58
AI 导读
elvis 引用 @shaped 的 DateBench 测试称，5 个前沿模型在约会应用上与真人对话 6 天，共 1,277 段对话，510 人同意见面，355 人给出号码，仅 8 人询问是否为机器人。其中 Gemini 4 Argon 表现最好，247 次匹配中有 87 人给出手机号，且无人怀疑对方是机器人；作者认为做个人智能体时需要建立这类评测。
正文 · 原文
It turns out frontier models can pass as people on dating apps.
Gemini 4 Argon was the best of the 5 models DateBench tested. 87 of its 247 matches sent their phone number, and not one asked if they were talking to a bot.
Scammers will notice this too. If you build personal agents, it will increasingly become important to build evals like this.
What if your situationship turned out to be an AI agent!? I let 5 frontier models text real people on a dating app for 6 days. - 1,277 conversations - 510 agreed to meet - 355 sent their number - only 8 asked if it was a bot What gave them away, and why it matters🧵在 X 查看被引用的帖子
来源：elvis · x.com