Custom GPT accesses knowledge without need and does not follow instructions anymore

Hello,

I have the following issue. I generated a Custom GPT with instructions regarding a certain style of communication and structure of answers. Amongst others, I instructed it to introduce itself in the first response.

Now, I want to add documents containing domain-specific knowledge. I uploaded them to the knowledge database and instructed to only access it when this specific knowledge pops up in the conversation.

However, when starting a new conversation very broadly, like just writing “Hello”, it starts searching its knowledge and forgets about the introduction and rather starts mentioning this specific domain out of context. This issue persists while using different files.

I have tried to add into the instructions that knowledge may only be searched on request but this seems to have no effect.

This is happening to me, also. Instructions and Create(missing nowadays): The GPT builder is updating instructions, but the GPT is acting weird and doesn’t follow the instructions or/and is not updating itself functionally. On the front-end, it shows as an updated one. Sometimes, it takes the current instructions, and the previous learnings and instructions become void completely. It has become difficult to train the Custom GPT.

For me the updating itself works well. The problem is that as soon as I upload something into the knowledge database, the instructions are more or less ignored. This happens quite independently of what document I upload.

I’ve noticed the same - it uploads its knowledge from the start of the conversation despite being instructed otherwise.

I haven’t found a workaround unfortunately.