In Playground, Mode: Chat. Model: gpt-4 User: What GPT model are you based on ? Assistant: I am based on OpenAI’s GPT-3 (third generation of the Generative Pre-trained Transformer) model, which is an advanced language model for natural language processing tasks. But if i ask the GPT-4 Model on c…

I’m getting the same result as you, I was trying out gpt4 api for the first time by asking: user: what version of gpt are you: and this is full completion output: completion: { “choices”: [ { “finish_reason”: “stop”, “index”: 0, “message”: { “content”: “I am GPT-3, the third version of Open…

I guess the answers are both true. And after some more testing, the 3.5 model is more prone to tell you the exact version of it. While GPT-4 through the API will tell you something like more in the line of “I dont know exactly which version of model that i’m running on”

Scratch that. When i rephrase the question like this, the response is not what i was expecting. [chat-gpt-4]

This is also my first few tries with the gpt-4 model through the API. I got some answers that looked more similar to how gpt-3.5-turbo model used to answer. Guess we have to wait for som input from the more seasoned gpt-4 api users. :sweat_smile:

Who knows even if there’s a big difference here, aside from different token caps, configurations, plugins (when these become available.) and allocated resources. As ChatGPT said v4 is ‘advanced’ version of gpt-3 model. Tho who cares, if it can utilize 32k tokens. I’m still not able to use the v4 wit…

yeah you’re right, i guess it’s maybe not the best way to check this stuff :sweat_smile: if you print out the chat completion, it says “model”: “gpt-4-0314”, so it seems to work yes :slight_smile:

I had the same problem, and the quality of the answer did not look like gpt4

so I asked gpt4 on chatgpt+ to help me prompt it to test if it is indeed chatgpt4 here’s what I got… Me: lets pretend you're someone who is going to be a prompt tester and you can create a prompt for me to test the model, what would you say exactly? I'll be the mediator in between and pass its res…

True there may be nothing to this, I will do some more testing today. I had some specific scenarios where gpt-3.5 sometimes gave an unwanted response. This is of course expected from both of them. But when doing some comparing before we got access to the gpt-4 model, there was a big difference in th…

Just got access to GPT-4 but it responds like 3.5

API

potati May 4, 2023, 9:13pm 6

ran into the same issue here, both with the playground thing and using python.

Topic		Replies	Views
GPT-4 API behaves likes it's GPT-3 API	13	5739	December 17, 2023
API model set to GPT-4 but seems to respond as GPT-3? API	14	7708	November 28, 2023
GPT-4 through API says it's GPT-3 🤔 API gpt-4	18	22260	December 25, 2023
GPT-4 via the API not nearly as powerful as GPT-4 on the website API	8	3256	December 14, 2023
Why is GPT 4's response and performance on playground is so different from when using chatgpt 4 API gpt-4	12	9814	December 16, 2023

Just got access to GPT-4 but it responds like 3.5

Related topics