Closed noname01t1 closed 1 year ago
Automated: 👋 Thanks for reporting this issue. @acheong08 will review it shortly. In the meantime, please check to see if there are any similar issues already closed.
I currently don't have access to browsing and thus can't reverse on my own. Please provide a sample browsing request and the resulting stream. You can find this from F12
This issue is stale because it has been open for 2 days with no activity.
Hi @acheong08 ,
So I am not entirely sure what yalls approach is (using openai's official API / using sockets for their chat.openai.com / using browser to navigate there / etc.) but this is the POST request to https://chat.openai.com/backend-api/conversation
inside the chat.openai.com when you select GPT4 browsing:
{
"action": "next",
"messages": [
{
"id": "aaa23ff3-1dc4-42c7-8e31-5dc9abbed21f",
"author": { "role": "user" },
"content": {
"content_type": "text",
"parts": [
"What's the age difference between dua lipa and her boyfriend?"
]
}
}
],
"parent_message_id": "aaa1f5d0-a18a-4619-95f7-0fc0ecd7b1bf",
"model": "gpt-4-browsing",
"timezone_offset_min": -120,
"history_and_training_disabled": false
}
Would that help?
Thanks. That is exactly what I was looking for. Will work on this soon
Can you send the corresponding response?
I have no way of testing it but based on the original issue, the problems seems to be parsing
Even without access to it on the web interface, sending the request got me access (I think)
{
"message": {
"id": "0826472e-c9e3-498d-93ae-75b9aec7e8ff",
"author": {
"role": "assistant",
"name": null,
"metadata": {}
},
"create_time": 1684637522.627016,
"update_time": null,
"content": {
"content_type": "text",
"parts": [
"..."
]
},
"status": "finished_successfully",
"end_turn": true,
"weight": 1,
"metadata": {
"message_type": "next",
"model_slug": "gpt-4-browsing",
"citations": [
{
"start_ix": 356,
"end_ix": 366,
"metadata": {
"title": "apnews.com",
"url": "https://apnews.com/hub/joe-biden",
"text": "...",
"pub_date": null
}
},
{
"start_ix": 366,
"end_ix": 377,
"metadata": {
"title": "apnews.com",
"url": "https://apnews.com/hub/joe-biden",
"text": "...",
"pub_date": null
}
}
],
"finish_details": {
"type": "stop",
"stop": "<|diff_marker|>"
}
},
"recipient": "all"
},
"conversation_id": "65453023-2a02-47e0-8cf8-069b06759841",
"error": null
}
Parsing issues come from
{
"message": {
"id": "23b01f22-5c70-4e62-b398-0c6f3da28f04",
"author": {
"role": "tool",
"name": "browser",
"metadata": {}
},
"create_time": null,
"update_time": null,
"content": {
"content_type": "tether_browsing_display",
"result": "",
"summary": ""
},
"status": "finished_successfully",
"end_turn": null,
"weight": 0,
"metadata": {
"command": "search",
"status": "running",
"args": [
"latest news on Joe Biden"
],
"message_type": "next",
"model_slug": "gpt-4-browsing"
},
"recipient": "all"
},
"conversation_id": "65453023-2a02-47e0-8cf8-069b06759841",
"error": null
}
Done
Is there an existing issue for this?
What would your feature do ?
To support the latest model: gpt-4-browsing
Proposed workflow
Working when "model" is set to "gpt-4-browsing"
Additional information
Current error:
If I get enough time, I'll try to submit a PR.