I used a decent prompt and asked my GPT what else it needed to know to provide the outcomes I wanted. It asked for things like case studies and best practices, etc. So, I wrote a training response, referring to a GPT called Link Reader GPT, which I’ve used to capture content. However, the moment I call upon @Link Reader GPT, everything after the @mention is deleted, prohibiting it from running, not to mention all of the other training I provided is forfeit in the response.
What am I doing wrong? I’m using GPT4 and am licensed at the Teams level.
1 Like
Hi @dunham_john welcome to the community.
Can you either provide a screenshot or clarify how you are attempting to inject the @ function in your prompt?
Also, it is typically a good idea to draft prompts outside of the UI. It provides a lot of benefits overall, but one important one is that it doesn’t get deleted when ChatGPT encounters an error.
Ahh. OK. I got what you mean now. If you mention the GPT you want to interact with first, then drop your prompt, you typically get better results. Calling the GPT after the prompt is inconsistent. Sometimes nothing happens and sometimes the prompts get wiped out or truncated. If you want to drop the prompt first, place the mention at the end of the prompt rather than the beginning. That more frequently results in the entire prompt staying intact.
1 Like
This makes a lot of sense. Once I am done pasting into the Link Reader, does the knowledge it collected go into the chat I was previously training? Or would I need to copy the content from the called GPT back into the main thread and ask for a summary of knowledge to ensure it. I think I’m almost there… not quite yet, though.
1 Like
Also, thank you! I appreciate the help!!!
1 Like
The knowledge does stay in the same chat and you do not need to copy the information and resend it when you switch back. You can also call multiple GPTs in this way and collect multiple perspectives and apply them to your final chat output or “main thread”.
To test this out, but really just for fun, you can always @ mention your GPT and ask it something like: “You see that last response? Does that make any sense to you?”
The saga continues. The Link Reader GPT generated uber-generic info, so I asked it which links it read, based on my training. It said
"I don’t have the capability to directly access or browse external links, so I haven’t read any links. I generate responses based on a mixture of licensed data, data created by human trainers, and publicly available information. My responses are generated based on patterns in the data I have been trained on, not from live browsing of websites or links.
If you have specific content or details from a link that you would like to discuss or need help with, you can copy and share that information here, and I can certainly help analyze or provide insights based on that information!"
Does this mean the GPT I’m referencing doesn’t work or that something else is wrong?
1 Like
A few different things impact its ability to browse the internet effectively, but based on this response:
Your issue is likely related to the nerfed ability of direct link access. Notice the issue in the below screenshot and how to get around it.
This is a great help. I will what seems to be an inferred search for a link now. Thank you!
