Errors after June 6 when accessing ChatGPT

API
, ,
1

After the news of July 6, I started having problems with the API.
Almost every answer ends with: I'm sorry, but I can't generate that story for you. or The server is overloaded or not ready yet
Anyone faced the same problem? Until June 6, everything worked :upside_down_face:

2

Welcome to the forum!

Can you post your API calling code? That does not look like a typical OpenAI error.

3

You have two separate issues it seems:

  1. you’re submitting unacceptable prompts that the AI is trained not to answer;
  2. “the server is overloaded”

Mitigations:

  1. You can pass the prompt input to the safety moderation endpoint first, and get a score that you can reject yourself instead of sending the policy violation to the AI and ultimately getting warning emails or terminations from OpenAI for unacceptable patterns;

  2. You can handle server-side problems yourself, either report the situation and ask the user if they want to retry, or retry a few times yourself while updating the user, even submitting to different gpt models for a fallback. Your code should be tolerant and handle the many types of API errors and HTTP reasons that can be returned.

4

It seems to me that the reason for my errors is GPT-4, which is now available through the API. 2 days ago with the same code everything worked