aifollow.news
Back Rohan Paul
Rohan Paul· @rohanpaul_ai · X· · Original publication time AI score54

Carnegie Mellon paper: Showing concrete choices helps shopping and travel assistants find more acceptable options than asking preferencesMachine translation

AI introduction

The research says users may not state in advance that they would accept a lower rating for a lower price, but may make that trade-off when they see actual options. With the same planner, adding one hidden trade-off to a request raised the share of plans accepted from 12% to 55%. The source also says current agents that mostly asked questions found at most 32% of plans users would accept.

Original post · AI translation

@rohanpaul_ai
Translation · refer to the original

卡内基梅隆大学的一篇新论文发现,负责规划旅行和购物的 AI 助手,与其询问人们的偏好,不如向他们展示几个真实选项,这样能找到更多人们愿意接受的选择。

像“评分为 3.8 或以上的餐厅”这样的旅行需求,会排除一家评分为 3.7、价格更低,而用户其实很乐意选择的餐厅。人们很少会说自己愿意为了省钱接受较低的评分,因为只有看到实际选项时,他们才会意识到这一点。

> 在使用同一个规划器的情况下,把这 1 项未明说的权衡加入需求后,用户接受的方案比例从 12% 提高到了 55%。

> 当前的智能体(Claude Code、Gemini CLI 等)大多只提问,很少展示方案,因此最多只能发现用户愿意接受的方案中的 32%。

> 展示方案的智能体明显比只提问的智能体表现更好。

当你的智能体将需求转化为方案时,应在确定需求规格之前,尽早向用户展示几个超出其明示限制的选项。

Source: Rohan Paul · x.com

Research
Found an error? Send a correction