#what are best practices for custom GPTs?
1 messages · Page 1 of 1 (latest)
I too am interested. It does seem to have some issues with PDFs, though it is capable of converting to readable formats
Well, there are size limits on the documents so if it's large you're probably better off using plain text format.
I would say plain text and slightly better markdown. but its a trade off I find that any docs always affect the prompt negatively so i try and not add them. there are people doing some crazy stuff but im not convinced...
I have the feeling as well that markdown would work better than txt. No idea about word docs.
Do you know if it is better one long document, many small, or makes no difference?
txt is the best choice, unless there are a lot of entries to your work, then PDF because ideal because it has indexing. I asked ChatGPT:
(short answer)
For CustomGPT applications that prioritize ease of processing and simplicity, .txt is often preferred. But for situations where document structure, navigation, and formatting are critical, .pdf can be the better choice.
Is anyone aware of any concrete testing that's been done between file content formatting and file types?
If not, I would love to collaborate with one or two other people on a testing research project that we can publish 2 gt 1 heads
There are several experiments I can think of that would tell us a lot.
DM me if I don't respond to this and you want to collaborate
I would be interested in running some test experiments although I am a bit short of time atm. What experiments do you have in mind?
pdf files can often be corrupted (even if you can still open and read them). txt is much better for templates. md is a lot better if you want to add subroutines/ alternative protocols. json and csv are great for datasets. zip is great for custom python modules.
PDF's are also very difficult to parse, so I wonder where the limits are, those sorts of things.
be careful with the information its quite easy to get the uploaded info from gpt, so give it good instructions or dont put sensative info into it 🙂
Definitely, I am all too aware lol
Does anyone know / is there documentation about how zip files are processed by a GPT, technically speaking? For example, are directory structures within the zip file respected/understood by a GPT that has one uploaded to it?
Further, is it know if the zip file does contain Python packages as directories with init.py files, can the GPT reason about code base structure?
I haven’t had any use cases for these ideas yet but I want to get a sense of the boundaries and edge cases
It's just unzipped normally. The files you uploaded are in mnt/data so by default the unzipped folder should appear right next too them.
Just tell ChatGPT the zip folder contains a python module and it'll acknowledge that.
Interesting, I’m going to have to experiment with this
https://chat.openai.com/g/g-ljZWFeJPb-basher-gpt-your-bash-linux-shell-tutor
use > printenv and it'll show you some bin folders. If you want Chat to actually use the modules, you might have to specify that you want it install it in /usr/local/bin.
I made this GPT for poking around the linux box that it does Advanced Data Analysis in. (Also, you can edit the env a bit and install your own programs but they do get wiped every now and again so you'll have to reinstall them or add a command for reinstall)
lmk what you think or if you need help
Do you know if the code runs in a micro-container or something to that effect? Something that can spin up very quickly or at least be provisioned very quickly on demand?
This is the first time I’m learning of the system internals of the code interpreter functionality. I imagine it’s a very locked down container
kubernetes
It doesn't have internet access
I've tried manually turning it on, but nope they got it under lock and key
Kubernetes is just the container orchestrator, I’m wondering what the container substrate itself is made from
This makes sense
ubuntu linux
That’s just the container OS, not the container system
oh that's what you meant
I'm not really sure