[NSFW] Do not reflect the contents of the attached document

  • Resolved yasunobu

    (@yasunobu)


    7 months ago

    I’m using this plugin with OpenAI assistant. And I’ve attached 5 word files. In the files, there is the statement about how to access a clinick.

    If I use Playground with this assistant, it works well. But from Chatbot dialog which is started from the ChatGPT chatbot aikon, it doesn’t work. Just say following messages.

    “We are sorry, but the file you uploaded did not contain information about Kochi Vein Care Clinic. Additionally, access to external websites is restricted, so it is currently not possible to search for information directly on the Internet.”

    How do I avoid this.

    The page I need help with: [log in to see the link]

Viewing 3 replies - 1 through 3 (of 3 total)
  • davejbriggs

    (@davejbriggs)

    (I’m another user not support).

    I also noticed this. My solution was to clone the assistant, and then change the bot assistant id in wordpress.

    In the API/Model tab, i also put in the assistant description again (not sure if needed but thought i would anyway).

    (My bot description expressly mentions that the files should always be consulted).

    From the plugin end, my thought is that there is an issue with the way that Threads are kept open, and that we either need to close a thread after X number of messages, or there should be a time out where a thread closes after X amount of time.

    From the OPenAi side – I noticed that if an assistant starts by refering to the files, it will continue to do so. If, however, it doesn’t refer to the files on its first answer, it is unlikely to do so again in that thread.

    Plugin Author kognetiks

    (@kognetiks)

    Hello everyone,

    The update to V2 beta may resolve this issue.

    Stephen

    Thread Starter yasunobu

    (@yasunobu)

    Thank you for sharing your approach. I tried the same method, but unfortunately, the situation did not change.

    I tested it by attaching a Word file that contains the clinic’s access information only, but still, the responses did not reflect the content of the text as expected.

    I’m beginning to suspect that ChatGPT might be programmed to avoid responding directly to specific personal information such as addresses, phone numbers, and email addresses. If I find a solution that works, I’ll be sure to share it here.

Viewing 3 replies - 1 through 3 (of 3 total)
  • The topic ‘[NSFW] Do not reflect the contents of the attached document’ is closed to new replies.