top of page
Search
Writer's pictureTim Kosmala

Troubleshoot ChatGPT in AutoResponder

Check out the general ChatGPT instructions here.


You keep getting a similar response even though you have changed the prompt? GPT also takes into account the reply history if used in the prompt. To fix the error, you can, for example, delete the reply history for this AutoResponder rule.


Your replies are cut off at a certain length? Increase the max_tokens parameter or tell the AI to write shorter answers.


If you receive the following error, this can have several causes.

Error: One of the parameters is invalid, your OpenAI account needs your attention or OpenAI is currently unavailable.

Please test the ChatGPT feature first with the default settings as described in this guide and as they are pre-filled by AutoResponder. If it still doesn't work, please don't give up.


Check ALL these steps please:

  • Make sure your API key is valid.

  • If you have used a Project API key and it doesn't work, please try to use a User API key (legacy) from here.

  • The OpenAI API only works in these countries. As a workaround you may use a VPN to connect from a supported country.

  • Make sure that you have enough credits here (OpenAI now asks you to buy credits in advance. Just start with a few dollars). (Note that the OpenAI API is not free to use like the ChatGPT web interface. But don't worry, it's very cheap. You can send around 25.000 AI messages for 1 USD with the default settings. Check the pricing.)

  • Make sure you didn't hit your maximum monthly spend (hard limit), otherwise increase it.

  • Your OpenAI account needs to be a member of an organization to use the API (skip this for now if you don't understand).

  • All parameters need to be correct as described in the GPT documentation (e.g. the model name, temperature setting etc.) (Don't worry, default settings are correct if you didn't change them).

  • Make sure the prompt+response combined doesn't exceed the supported token limit (but they are very high in the latest models).

  • When connecting an OpenAI Assistant, make sure the max_tokens are not set too low (try without a limit first). Otherwise, the execution will fail with an error.

  • Check your internet connection.

  • When using a different model, please check if you already have access to that model. Some require you to sign up for a waitlist on OpenAI. GPT-4 is currently only accessible to those who have made at least one successful payment of 5 USD through the OpenAI developer platform (you cannot use it with free credits).

  • In case you face a problem when using a fine-tuned model, please contact me via mail.

Pay special attention to your OpenAI credits here. Without credits, it won't work even if you have checked all other steps.

33,045 views

Related Posts

See All
bottom of page