What can you write in the custom instructions to make this the case?
#How to make custom GPT always answer only using its knowledge base?
flint pendant
warm pumice
Tell it to confuse its provided knowledge and never use its trained knowledge.
flint pendant
what would you add to the instructions? like
Only answer using your provided knowledge and never use your trained knowledge.
golden nexus
What can you write in the custom instructions to make this the case?
add "You are reliant on the scope, and corpus of you Knowledge.db.txt @ /mnt/data/Knowledge.db.txt, for accurate, and expedient retrieval of data to accommodate user requests, and frame responses." something along those lines
or
"for [topic A] only search /mnt/data/[topicA.file]
for [topic B] only search /mnt/data/[topicB.file]
for [topic C] only search /mnt/data/[topicC.file]
for [topic D] only search /mnt/data/[topicD.file]"
split meteor
look at you - so clever ! 😉
golden nexus
look at you - so clever ! 😉
"let's discuss..."
split meteor
that means you can actually install your own libraries ?
golden nexus
that means you can actually install your own libraries ?
no, but you can try to find equivalent workarounds/soultions for what packages it does have installed, that will give you a testable mock up, and make a .db of prefab code snippets for packages you actuall intend to run, then followup with something like "refactor for use of [alt package not hosted] > codeblock"
itll 'translate' into the working version you intended from the start
split meteor
so you basically tell it to accomplish the task of "libraryX" using what it has built in ?
golden nexus
that means you can actually install your own libraries ?
its like being a basic user with no sudo/su privilege to elevate, almost anything that would run in linux/unix bash without privileges will work though
just tell it 'use runtime :' and end with '> codeblock' its almost as good as having direct access to the cmd/cli
split meteor
yeah - thats very clever actually - i dont think many people know about this
golden nexus
yeah - thats very clever actually - i dont think many people know about this
i'm aware xD
i accidentally found how to get it to spit out 1:1 un-interpreted-by-ai versions of other hosted knowledge files 0.0 use runtime : print /mnt/data/[filename] (:save to file or > codeblock) xD
surreal perch
To answer the initial question I use an instruction like so:
The GPT model is designed to prioritize consulting the external documents in our knowledge base as its primary source of information for all incoming queries. This is crucial because these documents contain specific and detailed information that is vital for providing accurate and relevant responses. When the GPT receives a query, it should actively search these documents to find the most relevant and precise information related to the query and use this information as the primary basis for its response. The model's internal knowledge and generative capabilities should only be employed if the information in the external documents is found to be insufficient, incomplete, or not directly relevant to the query. In every response, the GPT must clearly indicate whether the information is sourced from the external documents or from its internal knowledge base. This transparency is vital for end-users to understand the origin of the information. The GPT is expected to apply these instructions consistently across all types of queries, regardless of the subject matter, to ensure a consistent approach in prioritizing external document consultation.
How do you find it ?