Cheating with model providing?

API
1

Hi there!

I’m totally newbie so my question can be stupid, sorry for this.
But could you look at this please?
I’m asking in my API-request:
What's model you are: GPT 3.5 or GTP 4?

And got:

{
  "model": "gpt-4o-2024-08-06",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "I am based on the GPT-3.5 architecture.",
        "refusal": null
      },
}

Yeah, I see in the response “model”: “gpt-4o-2024-08-06” but AI doesn’t know anything about it???

Thank you!

2

that is correct, the model does not know what model it is, it is not in the training data for some reason.

according to @anon22939549 “The model information cannot be in the training data because the training data needs to be curated before the model can exist after being trained on the training data.”

solution should go to him, my response was written when I was really tired

2 Likes
3

I see. But when you’re asking the exactly the same question on chatgpt[dot]com via the same “gpt-4o” model (which I use for API requests) you got the answer:

I’m based on the GPT-4 architecture

How to explain it? =/

4

The ChatGPT model has a system prompt from OpenAI baked into every interaction giving it information like the current date and the fact that it is “based on the GPT-4 architecture”.

The only reason the model knows is because it was told

1 Like