latent space 上有一篇文章讲到基于O1 类推理型模型的产品设计思路,总体思想是产品并非chatbot ,而是“ report generator” ,由此,文章认为有三点值得关注的:
1. 展示推理型模型回复的层次
2.让复杂的信息层次更容易浏览
3.让用户容易管理上下文的具体内容
1. Make it easier to see the hierarchy of the response (think a mini table of contents)
2. Similarly, make the hierarchy more easily navigable. Since every request is usually larger than the height of the window, I would take a Perplexity like approach where each question/answer page gets a section vs. freeform scroll. Within an answer, things like sticky headers, collapsible headers, etc. could really help)
3.Make it easier to manage and see the context you’re providing to the model. (Ironically Claude’s UI does a much better job of this — when you paste in a long piece of text, it renders as a little attachment). I also find that ChatGPT Projects don’t work nearly as well as Claude’s, so I’m copying and pasting stuff a lot.
原文:
www.latent.space