Inception

Inception is a project primarily written by OpenAI's ChatGPT. Its purpose is to demonstrate the capability of ChatGPT to generate code for various tasks with minimal human intervention.

The project consists of a set of Python scripts and tools for introspecting and sending code to ChatGPT, as well as for decoding and writing the code generated by ChatGPT.

Current Capabilities:

Sending code files or directories to ChatGPT using the isend command.
Decoding and writing files received from ChatGPT using the idecode command.
Automatically generating function signatures for Python files.
A Docker setup for easy deployment and testing.

This project is still a work in progress, and ChatGPT has contributed the majority of the code.

Usage Example

The process of creating this README is an interesting usage example of sorts.
Here's an example of how one might use Inception to write a new app and then modify it slightly:

First, run isend hello_world.py to read the existing code.

[ .. user runs `isend hello_world.py` to read existing code ..]
USER:

[Reminder.md](prompts/Reminder.md)
# CONTENT
## WRITE THIS FILE hello_world.py
import random

def print_hello_world():
    print("Hello, World!")

if __name__ == "__main__":
    print_hello_world()
# NO MORE FILES FROM CHATGPT
<example output of `inception-function-sigs`>

You can see my existing code above.  My first task for you is to modify this script to print its message in one of 3 different random languages each time it's called.

Don't forget, I can't understand code unless it's in a content block!

ChatGPT responds with a new content block:

ChatGPT:

```
# CONTENT
## WRITE THIS FILE hello_world.py
import random

def print_hello_world():
    greetings = [
        "Hello, World!",  # English
        "Hola, Mundo!",   # Spanish
        "Bonjour, le Monde!"  # French
    ]
    greeting = random.choice(greetings)
    print(greeting)

if __name__ == "__main__":
    print_hello_world()
# NO MORE FILES FROM CHATGPT
```

The user pastes the content block into idecode.

The user issues git diff locally to see the changes.

[... user issues `git diff` locally ..]
<example output of `git diff`>
diff --git a/hello_world.py b/hello_world.py
index 0a0b1c2..3d4e5f6 100644
--- a/hello_world.py
+++ b/hello_world.py
@@ -1,5 +1,11 @@
 import random

 def print_hello_world():
-    print("Hello, World!")
+    greetings = [
+        "Hello, World!",  # English
+        "Hola, Mundo!",   # Spanish
+        "Bonjour, le Monde!"  # French
+    ]
+    greeting = random.choice(greetings)
+    print(greeting)

 if __name__ == "__main__":
     print_hello_world()

TODO

Checkpointing with git: Automate git check-ins and add functionality to squash commits from a previous defined point once a change is somewhat fully baked. ChatGPT should be able to come up with the commit messages and the squash's.
Direct communication with OpenAI API: Implement a system that starts with GPT-3.5 and then attempts to talk to GPT-4 or prompts the user to manually send the task to GPT-4 if necessary.
Send and/or receive code changes: Improve the process of sending code changes to ChatGPT by finding a more predictable way to present changes, reducing code duplication.
Ask ChatGPT to decide what sections of code to send for context: Develop a system that uses a separate GPT instance to figure out which code matters to the tasks at hand, making the process more efficient.
Add unit tests: Enhance the project by including unit tests for the various features and functionality.
Lean on TDD for error checking: Rely more on Test-Driven Development for error checking, allowing ChatGPT to write tests, [manually review them,] then write the code, return failures, and iterate on the solution with minimal human intervention.

jaredj/inception

Inception

Usage Example

TODO