[Request] 对话分支

[naimehao]

🥰 需求描述

现在可以修改对话中之前的提问，然后重新生成，但此前的提问就找不到了，而且生成的内容并没有覆盖之前AI的回复，而是在之前AI回复之下又生成了新的回复，这会让人比较困惑。

举例：
比如我问AI，xxx的方案。
AI给了两个，我两个都想具体了解，但又不想在一次会话中掺入与最终方案无关的支路对话内容，就会进行分支。
我第一次会说，我想了解A方案，完了我会将“我想了解A方案”修改为：我想了解B方案，然后我自行判断哪个更好，哪个更好，就沿着哪条路走下去，甚至我在A方案中走了10步后，我想想，还是B方案好，我再回到最开始“我想了解A/B方案”那个分支的地方，再沿着B方案走下去。

我知道现在可以对自己的和AI生成的内容进行修改和删除，但是修改了之后，不会覆盖之前的内容，而是在下面生成了新回复，看上去就比较奇怪，而且我也不知道这是不是会带到上下文中去。

还有就是对于修改AI回复的这个功能我持怀疑态度，这样确实很自由，但很多时候，AI行就是行，不行就是不行，如果AI回复的内容中部分不正确，你改对了，但AI本身能力不行，你这里改了，后面该错还是错。

🧐 解决方案

完全仿照Claude，可以左右切换对话分支。（甚至我一直觉得Claude可以针对该特性搞个树状图的可视化出来。）

我是Claude这两天被封了之后开了你们的云版本的会员，这个功能我觉得可以搞成云版本特供。😆

📝 补充信息

No response

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

🥰 Description of requirements

Now you can modify the previous questions in the conversation and then regenerate them, but the previous questions cannot be found, and the generated content does not cover the previous AI reply, but generates a new reply under the previous AI reply. This It will be more confusing.

Example:
For example, I asked AI, xxx’s plan.
AI gave two of them, and I wanted to know more about both of them, but I didn’t want to include branch dialogue content that had nothing to do with the final plan in one conversation, so it would branch.
For the first time, I will say, I want to understand plan A. After that, I will change "I want to understand plan A" to: I want to understand plan B, and then I will judge by myself which one is better, and which one is better, and I will follow that path. As the road goes on, even after I have taken 10 steps in Plan A, I think about it, Plan B is better. I will go back to the branch of "I want to understand Plan A/B" at the beginning, and then follow Plan B. Go down.

I know that I can now modify and delete my own and AI-generated content, but after modification, the previous content will not be overwritten, but a new reply will be generated below, which looks strange, and I don’t know this. Will it be brought into context?

Also, I am skeptical about the function of modifying the AI ​​reply. This is indeed very free, but in many cases, if the AI ​​is good, it is good, and if it is not, it is not good. If part of the content of the AI ​​reply is incorrect, you have corrected it, but the AI If your own abilities are not good enough, if you make a change here, it will still be a mistake later.

🧐 Solution

Completely imitating Claude, you can switch dialogue branches left and right. (I have always thought that Claude could create a tree diagram visualization for this feature.)

I am a member of your cloud version after Claude was banned for the past two days. I think this feature can be made special for the cloud version. 😆

📝 Supplementary information

No response

[lobehubbot]

👀 @naimehao

Thank you for raising an issue. We will investigate into the matter and get back to you as soon as possible.
Please make sure you have given us as much context as possible.
非常感谢您提交 issue。我们会尽快调查此事，并尽快回复您。 请确保您已经提供了尽可能多的背景信息。

[SpeedupMaster]

#4616

[lobehubbot]

✅ @naimehao

This issue is closed, If you have any questions, you can comment and reply.
此问题已经关闭。如果您有任何问题，可以留言并回复。

[arvinxx]

@naimehao 这个月就会上，而且你说的树状图模式未来也会有。这些也都会全开源

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

@naimehao It will be available this month, and the tree diagram mode you mentioned will also be available in the future. These will all be open source

[arvinxx]

还有就是对于修改AI回复的这个功能我持怀疑态度，这样确实很自由，但很多时候，AI行就是行，不行就是不行，如果AI回复的内容中部分不正确，你改对了，但AI本身能力不行，你这里改了，后面该错还是错。

这个不会的，事实上这个用起来还是很舒服的。主要是改小细节。比如 import abc from './yourpath' 这样的输出，我直接编辑之后改成 import abc from './index.ts' ，这样继续对话时就会保持这个 import abc from index.ts 写法。不用再去说把 yourpath 改成 index.ts 然后等重新输出一遍，这个体验是好太多了。

而且修改模型输出，本质上是在使用 few shots 的方式来规范模型输出。我个人一直认为 few shots 应该是所有提示词工程中最有效的方式。

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Also, I am skeptical about the function of modifying AI replies. This is indeed very free, but in many cases, if the AI ​​is good, it is good, and if it is not, it is not good. If part of the content of the AI ​​reply is incorrect, you have corrected it, but The AI ​​itself is not capable. If you change it here, it will still be wrong later.

This won’t work. In fact, it’s very comfortable to use. Mainly to change small details. For example, for output like import abc from './yourpath', I directly edit it and change it to import abc from './index.ts', so that this import abc from index.ts will be maintained when the conversation continues. This realization.

And modifying the model output is essentially using few shots to standardize the model output. I personally have always believed that few shots should be the most effective method of all prompt word projects.