Web chat default temperature for GPT 3.5 and 4

ChatGPT
1

What is the default setting of the temperature for GPT 3.5 and 4 on the web version? I see numbers like 0.7, 0.8 and 1 on the internet, but they seem to be guesses.

1 Like
2

I’m curious about this as well. I’m using CodeGPT for VSCode and the default temperature setting there is 0.3, but I’d like it to match the ChatGPT web UI since that’s been really useful and pretty accurate.

3

Same thing here. How does this forum works? Thousand of people asking questions and nobody answering them? :slight_smile:

4

I can only say that for the Web interface it is NOT 0.0

I was getting slightly better results on web vs API and since the model is supposed to be the same (“gpt-3.5-turbo” as of 2023-05-17) then “temperature” could be the cause.

I changed it from 0.0 on the API calls and results were more aligned with Web results.
Which value? I don’t know, but not 0.0

1 Like
5 
import openai
openai.ChatCompletion.create(
  model="gpt-3.5-turbo",
  messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "How do I change the temperature setting?"},
        {"role": "assistant", "content": "To change the temperature setting, you can include the 'temperature' parameter in your API request. **The default value is 0.8**."},
        {"role": "assistant", "content": "For example, you can set the temperature to 0.5 to get less random output."},
  ],
  temperature=0.5,
)

I pretty sure its 0.8 default for web chat.openai. I found it through api call script

1 Like
6

For the API the default value is 1 according to this source:

7

I’d like to inquire about the temperature configurations as well. I’ve noticed that it produces a greater number of outcomes.

8

Hey everyone, I found the solution for this and it was stated on the docs.
Just search this on Google: “chatgpt ?pivots=programming-language-chat-completions”

If you want the GPT-35-Turbo model to behave similarly to [chat openai com], you can use a basic system message like “Assistant is a large language model trained by OpenAI.”

{“role”: “system”, “content”: “Assistant is a large language model trained by OpenAI.”},
{“role”: “user”, “content”: “Who were the founders of Microsoft?”}

If you want more creativity and diversity, update the temperature to 0.08