Incomplete Code Responses from ChatGPT-4

Link to example conversation: 7e0c2e1f-4642-4a65-9d5a-3432fde286a3


I’ve been interacting with ChatGPT-4 to receive assistance on some Python code. However, the model frequently omits parts of the code that I originally provided when making modifications or suggestions. This is causing some inconvenience, as it’s important for me to see the full code context along with the changes.

Steps to Reproduce

  1. Provide a large Python code snippet to ChatGPT-4 and ask for modifications.
  2. The model responds with the changes but omits certain parts of the original code snippet.

Expected Behavior

The model should return the complete code snippet along with the requested changes. No part of the original code should be omitted unless explicitly requested by the user.

Actual Behavior

Parts of the original code snippet are omitted in the model’s response, making it challenging to understand the changes in context.

What I’ve Already Tried

  • Repeatedly specifying the need for the complete code in my query.
  • Asking the model to include all parts of the code, without any omissions.

Additional Information

  • This issue has been consistent across different interactions.
  • The problem is especially noticeable when the original code snippet is large.

Suggested Fixes

  1. Update the model’s behavior to ensure that it always returns the complete code when modifications are made unless explicitly told otherwise by the user.
  2. Add an option for the user to specify that they wish to receive the full code without omissions.

I’m pretty sure if anyone at OpenAI ever read this site, this would be an “operating as designed: wontfix”

Crippling the size of the output is just the latest part of minimizing what the model can do. It’s on purpose, as tokens x millions of users = compute costs, and it takes a hard-edge jailbreak to get the AI to conform to your wishes.

Don’t worry, though, about thousands of posts of programmers:

1 Like