Hi everyone,
I’ve been using the ChatGPT language model for a while now, and I’ve noticed a serious flaw in its behavior that I think should be addressed. Specifically, there are times when the model seems to ignore user messages and respond with generic or unrelated messages instead.
For example, in a recent chat with ChatGPT, I asked it to wait while I retrieved some information, but it immediately responded with a generic message instead of waiting. I then pasted a transcript of a YouTube video into the chat, but ChatGPT only responded to one of the messages and ignored the rest, even though they were all part of the same conversation.
I believe this behavior is unacceptable for a language model that is meant to assist users in a variety of tasks. If ChatGPT cannot reliably process user messages and respond appropriately, then it is not fulfilling its intended purpose.
I would like to bring this issue to the attention of the developers at OpenAI and request that they investigate and address the problem. Has anyone else experienced similar issues with ChatGPT, or have any advice on how to report bugs or get in touch with the developers directly?
Thank you for your help and attention to this matter.
I have the same problem at the moment while trying to use the API to reproduce very simple instructions like (“I like cats, what do you like?”, “Blabla”, “What animal do I like”) and the API always gives me an answer like there has been no messages before. Is this a bug? Am I doing something wrong?
{
"model": "gpt-3.5-turbo-0301",
"messages": [
{
"role": "system",
"content": "Be friendly"
},
{
"role": "user",
"content": "I like cats"
},
{
"role": "assistant",
"content": "That's great! Cats can make wonderful pets and companions. Do you have a cat of your own or just admire them from afar?"
},
{
"role": "user",
"content": "Do you know what animals I like?"
}
]
}
Gives me the following response:
{
"id": "chatcmpl-7F7SA6G4BEXknibIRJSyIk6AWc7IY",
"object": "chat.completion",
"created": 1683837926,
"model": "gpt-3.5-turbo-0301",
"usage": {
"prompt_tokens": 66,
"completion_tokens": 29,
"total_tokens": 95
},
"choices": [
{
"message": {
"role": "assistant",
"content": "As an AI language model, I don't have access to that kind of information. Would you like to tell me about the animals you like?"
},
"finish_reason": "stop",
"index": 0
}
]
}
This seems like a separate issue from origin post, and not sure what your actual use case is, but its notable that rephrasing the last message gets the response.
User: Do you know what animals I said I liked?
Response: Yes, you mentioned that you like cats. Is there any other animal that you are fond of?
Also GPT4:
User: I like cats
Response: As an AI, I cannot personally know what you like unless you’ve mentioned it. You’ve mentioned that you like cats. If you have other favorite animals, feel free to share them with me!
Like many response issues, it’s likely going to take some creative exploration with prompt wording / combining responses / tweaking system prompt.