#GPTs wrongly classified as "templates" (information disclosure)

1 messages · Page 1 of 1 (latest)

narrow atlas
#

[Some users have posted](#1173675836277276762 message) in various channels that they have received the ability to see and remix the inner workings of GPTs via the web UI. It appears as though this feature is intended to be used for "templates", however, there is no way for GPT makers to toggle whether or not their GPT is considered a "template".

Because there is no "template" option, this may lead to other users stealing other people's GPTs without permission and taking credit. The instructions inputted by the maker are their property and it's important that they have the ability to tell the GPT not to reveal the instructions and to make sure their GPT is not a "template" if they don't wish to "open source" it.

Thank you for listening to my Ted Talk.

image.png
nova sinew
#

I don't think the Instructions are someones property. It also promotes competition and therefore helps the entire community. There's only one reason people don't want to share their prompts - money. And I find it pathetic and ridiculous that people make such a fuss for the 50 cents they will get from OpenAI.

narrow atlas
#

I'm not concerned about money. The problem is that it enables (and encourages) other people to steal GPTs / ideas / prompts and not give credit. Have you ever noticed that companies don't copy the Coca Cola ingredients off the back of the bottle and make their own knockoff? They could easily do so as there are documented instances of people reverse engineering the recipe and steps, but the reason you don't see knockoffs is because the recipe still belongs to Pepsico.

ChatGPT is the ingredients, prompts are the recipe. Now if Sam forgets to put in a template toggle option I'm gonna scream! rage

nova sinew
# narrow atlas I'm not concerned about money. The problem is that it enables (and encourages) o...

No, you're looking at this from completely the wrong angle. It's in the name. Instructions are those instructions you give to your workers. The worker is GPT4. The ingredients are DALL-E, Bing, Code Interpreter, API's, PDF's, Word Documents, Images and many more things you can upload yourself. I can't download those ingredients, but I can see how you treat your workers and maybe copy that for better performance.

#

Or while we're at it... DALL-E, Bing, code interpreter and API's are more the tools that the worker has. Everything else is ingredients

narrow atlas
# nova sinew No, you're looking at this from completely the wrong angle. It's in the name. In...

If you've done any amount of prompt engineering, you'll know that's not how it works at all. Small changes in the prompt can lead to drastic changes in LLM behavior and complex GPTs require a lot of work to get working desirably. In a workplace, you just tell the person who's been there the longest to "go onboard the new hires". This is completely incomparable.

More importantly, take a look at the text in the image. This is clearly supposed to be a toggleable setting already. I'm not sure if they pressed a wrong button or what, but this is a bug by definition unless they somehow decide that they're going to make this intentional now

nova sinew
narrow atlas
#

You don't do well with analogies, do you?

nova sinew
#

I think I was pretty clear in explaining how to make this analogy logical