IPython + GPT-4

A quick implementation of a random idea: what if ipython had a magic function that sent the current session to gpt-4 along with a prompt, and automatically evaluated the result in the current session?

Installation

pip install ipython_openai

Then, you can either load the extension during an ipython session with %load_ext ipython_openai or add the following to your ipython config.

c.InteractiveShellApp.extensions = ['ipython_openai']

Configuration

Your OpenAI API key should be available in the OPENAI_API_KEY env var. You can configure the model used by adding an entry to your ipython config: c.GPT.model_name = 'gpt-4-turbo-preview'.

Usage

This extension adds two new "magic" functions documented below. Each one captures the current history of the ipython session and sends it to the OpenAI API along with the prompt given after the command. The result is streamed back to the session. Each invocation of the magic functions can also see the output from previous ones.

gen

In [1]: gen a 5 line greeter function with randomness
import random
def greet():
    names = ["Alice", "Bob", "Charlie", "Diana", "Edward"]
    greeting = ["Hello", "Hi", "Hey", "Greetings", "What's up"]
    print(f"{random.choice(greeting)}, {random.choice(names)}!")

Run it? (y/n) y

In [2]: greet()
Hello, Bob!

ask

In [3]: ask what happened
When the `greet()` function was called, the following occurred:
1. The function selected a random name from the list `names`, which contains five different names: "Alice", "Bob", "Charlie", "Diana", and "Edward".
2. It also selected a random greeting from the list `greeting`, which includes five different greetings: "Hello", "Hi", "Hey", "Greetings", and "What's up".
3. It combined the selected greeting and name with a formatted string to create a message.
4. Finally, the function printed this message to the console.
Since the selections are random, the exact output can vary with each call to `greet()`, displaying a greeting and a name randomly chosen from the lists provided.