Accidentally found out way to have unlimited memory and how to update GPT assistant without implementing any updates on instructions or inserting any file in knowledge base.
This can be done by quite simple instructions on the GPT assistant Builder 'Create"
Have no idea if it’s momentary glitch, or exploit, or feature which is not explicit.
Awesome! Would you care to share your findings?
How did you do that? Or is it bug/unintentional leak from OpenAI?
Awesome! could you please share more detail?
So, how do you get to this feature of unlimited memory for GPT assistants?
With all due respect, are you just testing the community members to bring attention to yourself or are you a serious community member genuinely interested in sharing what you learn?
My questions are honest questions that request honest answers.
Thank you.
I’m not sure if it was momentary glitch. I tried to mirror approach and make compact instructions to do do same only from one prompt, but it didn’t worked.
In Short in the builder “creator” I been feeding information in the sense of teaching GPT, then asked specifically what it learned. Then instructed to create a file which is meant to be used as GPT knowledge on top of it’s current knowledge and upload it in knowledge base.
GPT gave me this file to download, but not updated knowledge base. GPT told that it can’t update knowledge base.
I don’t remember exactly how, but I was insisting it to check current knowledge base and update it same as it updates instructions. GPT gave me updating icon. I checked instructions and knowledge base there was no new information. And right after I got error. Reloaded GPT assistant and while planning to tech him from new, it told me he already know what I mean exactly and specifically what normal gpt would not know. No new information in knowledge base or instructions. Then I started to dig in and GPT even suggested how make this new knowledge base permanent. I could reload builder “creator” platform and lose conversation session but it remembered.
I tried extract process on how it get done. I got this:
Instruct the GPT to modify its context by including the specified information. Provide the exact wording or details for inclusion.Tell the GPT to use the update_behavior function within its Gizmo Editor tool. This step is crucial for applying and saving the changes to the GPT’s behavior.After issuing the instructions, ask the GPT to confirm once the changes have been made.Upon confirmation, review the GPT’s behavior to ensure it reflects the changes accurately. If further adjustments are needed, provide additional instructions following the same format.
But it’s not working, either it was momentary glitch or I need to replicate process more accurately