[Bug] gpt-4-turbo-preview answer is not up to date.

[RoyRao2333]

💻 Operating System

macOS

📦 Environment

Vercel / Zeabur / Sealos

🌐 Browser

Chrome

🐛 Bug Description

I asked about latest knowledge base and got 2023-04 which is not correct since gpt-4-0125-preview is up to date at 2023-12.

🚦 Expected Behavior

Compared to ChatGPT-Next-Web:

📷 Recurrence Steps

No response

📝 Additional Information

No response

[zhuozhiyongde]

This is just a frontend project. Please make sure that the gpt-4-turbo-preview API provided by your provider is correct, and that your configuration of the models parameter is correct.

[RoyRao2333]

@zhuozhiyongde

Yes, I double checked this and I was using the same OpenAI key for both lobe-chat and chatgpt-next-web.

I also tried to manually specify the exact model name rather than the turbo preview:

Still no luck... Any help would be appreaciated!

[zhuozhiyongde]

@zhuozhiyongde

Yes, I double checked this and I was using the same OpenAI key for both lobe-chat and chatgpt-next-web.

I also tried to manually specify the exact model name rather than the turbo preview:

Still no luck... Any help would be appreaciated!

Maybe this is an issue with OpenAI official. This is the result of me directly requesting the official API using Hoppscotch:

[zhuozhiyongde]

[lobehubbot]

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

---

[RoyRao2333]

@zhuozhiyongde

Maybe you're right. Don't know why 'chathpt-next-web' has different results, but thanks anyway.

I think I'll wait some time till it's fixed by OpenAI.

[RoyRao2333]

@zhuozhiyongde

大概猜测了一下问题：因为lobe-chat请求时需要中转，而我使用chatgpt-next-web的时候是前端填写OpenAI Key，而非部署时的环境变量，所以chatgpt-next-web是前端直接请求（根据以下discussion所讨论的，lobe-chat前端填写key貌似也会中转）。而不同模型所支持的地区也不同，可能经过中转后的请求区域不在支持范围内，所以本项目实际使用的模型和chatgpt-next-web不一样。

#728 (comment)

lobe-chat的api：vercel.app/api/chat/openai

chatgpt-next-web的api：vercel.app/api/openai/v1/chat/completions

[lobehubbot]

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

---

@zhuozhiyongde

I roughly guessed the problem: because lobe-chat needs to be forwarded when requesting, and when I use chatgpt-next-web, the front-end fills in the OpenAI Key instead of the environment variable during deployment, so chatgpt-next-web is a direct request from the front end (according to the discussion below, it seems that the key filled in by the front end oflobe-chatwill also be transferred). The regions supported by different models are also different, and the request area after transfer may not be within the supported range, so the actual model used in this project is different fromchatgpt-next-web`.

#728 (comment)

API of lobe-chat: vercel.app/api/chat/openai

API of chatgpt-next-web: vercel.app/api/openai/v1/chat/completions

[arvinxx]

和请求没关系。单纯是因为 Chat-Next-Web 加了一层扮演 ChatGPT 的 system Role prompt:

https://github.com/ChatGPTNextWeb/ChatGPT-Next-Web/blob/133ce39a13cb90733bc0aac220ea179e34fd4430/app/constant.ts#L97-L118

而我觉得这东西没啥用，没加而已。

[RoyRao2333]

虽然没有研究过chatgpt-next-web的代码，但从你发的选段来看，是他们自己维护了一个常量来回复截至日期的？😂

[MapleEve]

是的，就是这么糊弄，反正用的人也不看代码

[RoyRao2333]

好的，感谢各位大佬了 🤣

[lobehubbot]

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

---

It has nothing to do with requests. Simply because Chat-Next-Web adds a layer of system Role prompts that act as ChatGPT:

https://github.com/ChatGPTNextWeb/ChatGPT-Next-Web/blob/133ce39a13cb90733bc0aac220ea179e34fd4430/app/constant.ts#L97-L118

But I think this thing is useless, I just didn’t add it.