[March 2024]: Summary of problems that STILL exist on GPTs/GPT dev

My GPT was one of the earliest released to the public. I wrote a post about how I migrated my plugin to GPT about 4 months ago, 2 days after the announcement of GPTs. I believe that I should have some domain knowledge in terms of GPT building and I’d like to represent for other developers today. I’m always hopeful that sharing my insights could help foster the dev community.

GPTs today are not reliable or pleasant to use. It’s not friendly for GPT developers either. Here are the list of problems that I still encounter today in March 2024.

  1. Repeated sign-ins within 1 hour, sometimes even after 10 mins.
    If you use OAuth for your GPT actions, the access tokens and refresh tokens that are returned from token_url are not stored / used by GPT properly. The access token from my server should expire in more than 1 hour, so if ChatGPT had kept the access token and used it, it would’ve still been able to make API calls without refreshing tokens. However, it’s not the case.

  2. Flakiness in OAuth
    After user is redirected back to ChatGPT through oauth, there is a flaky error that shows gpt-4-plugins at front and loginSuccess=false by the end of the browser url. This does not happen all the time but I have witnessed for a few times. Fortunately, my customer captured it in a screencast and I’ve uploaded it YouTube (link).

  3. GPT inaccessible or not found
    This issue was brought in Jan 11th by this post and it is still occurring as of 03/03/2024.

  4. Network call errors for no reason. I have confirmed through my server logs that it is not an issue on our server because our server did not even receive any requests upon the time when I asked the question. I believe that something prevented ChatGPT from sending a network call to our server.

  5. Privacy settings sometimes disappear and thus we can’t log in or log out through “Connected accounts”.
    I posted about this yesterday. I thought that it’s an edge case that would only happen to my customer. However, I encountered the same error as well.

The problems above have caused tremendous overhead for both users and developers. Of course, there is a chance that I am missing something and I could fix them on my end. Happy to hear about your advice. However, I still wish that my post could attract attention from the OpenAI team. I am kindly asking the team to take care of the issues mentioned above, if possible.