Editing files directly in my gpt

Hi everyone,

I build custom GPTs for fun and various artwork/story projects. I usually do this by making a custom gpt and uploading its knowledge base of things it needs to know for the story (characters, locations, the synopsis, etc.)

Whenever I update the story or adjust acharacter’s appearance, I have been editing the file directly, uploading the new file to my GPT’s knowledge base and deleting the old file. This gets pretty tedious and would like to know if anyone has tried anything like this and has a better workflow?

Also, is there any way to tell my custom GPT to edit the file directly and update it in its knowledge base? I have tried doing it before and it doesnt seem to work, but maybe its my prompting…

thanks!

That sounds like an interesting idea, and a possible source of AI-created data loss when the AI goes rogue, but there is no such mechanism.

A similar compartmentalization of user data as you might need is the cross-session memory tool of ChatGPT, but there is no version that is specifically memory for one GPT for one user.

GPTs are made to be shared. Imagine a user of your GPT says “I don’t need any character sheets any more, delete everything”. No, it isn’t going to work that way.

GPT files are presented to the AI only in the form of a file_search function it can employ.

2 Likes

Your case exhibits a common or generic functionality that is domain-agnostic. It’d be nice to have this functionality natively supported. A context needs to be modifiable, refreshed… overtime.

I use almost nothing but personal GPTs like this. Personally, I think this is the future of the CustomGPTs. Public GPTs are very hard to position strategically in a profitable way that are universally helpful.

Unique instances of CustomGPTs for content creation are a brilliant use. (And I think very applicable to small businesses, commercially.)

There is no way to have your GPT edit it’s own files, nor would you want it to for this type of creative work. The AI hallucinates too much to be effective at keeping your story’s framework true to your vision without your supervision.

So you have to update it’s files, which means you need more efficient ways of working with your characters.

In my own GPTs I’m having a lot of success with Character Mockups. Detailed character background stories, stats, pictures—a’la cartoon model sheets—that give the AI the story rules by which the character is governed. From this mockup, which you can save as a PDF, the AI can generate a more-consistent character. So, instead of you updating it with every little change, you have a more versatile document to work with that only requires periodic updates.

From there, since this is a PersonalGPT, you have a lot of immediate options with connecting it to, say, Google Drive using custom Actions that are difficult to manage if you want your GPT to be public. (You create a connection for yourself, and only yourself, using the native Google oAuth and API. Not “easy,” but secure, and pretty awesome, tbh.)

Here, you can give your GPT far greater control over data. So, for example, you can load a Google Sheet of Monsters into the Data Analyzer in your GPT via the AI Action, work on that list, then connect it directly back to the Sheet in Google Drive. Similarly, you can create flows for summarizing conversations and create a running dialogue.

Here, a word of caution. AI works with a Context Window. Stories, world building, require a TON of context. You have to become good at summarizing your own work in a way that meaning is maintained, which is why you can’t put your content projects on full-auto.

Longform storytelling is one of AI’s greatest challenges.

2 Likes