The new Custom Instructions are great and I can already see how it will help quickly spinning up new conversations without starting from scratch each time. One thing that would make it even better is if we could have a number of saved custom instructions.
For example, I’m a software developer and at work we have one set of libraries and patterns we prefer, while at home I have a different set. At home, I also use ChatGPT for non-coding situations like creative writing for D&D.
It would be a massive boon if I could save my “WorkCoding” custom instructions, my “HomeCoding” custom instructions, and “D&DWriting” custom instructions and when starting a new prompt I could select these from a drop down.
I would add to this and say it’d be helpful to also have some built-in prompts for custom conversation profiles. For example, I regularly prime my work-related conversations with the same context. It’d be nice to just select a conversation type, and that info would already be loaded into the conversation. Or, being able to select from a library/menu of commonly-used/saved/custom prompts.
I would appreciate if the custom instructions were given a larger character limit than 1500 characters because some contexts require detailed information and explanations.
The new Custom Instructions are fantastic. I’m a software developer ，i have to deal with programming issues, open-llm training and inference problems, Linux environment and conda environment, and some writing issues. I need at least five different Custom Instructions and switch between them quickly. I hope this new feature will be available in the next version!
I completely agree. I recently unsubscribed from GPT Plus because it lacks this feature. Unfortunately, I had to switch to Claude 2 with TypingMind, which does have it.
Custom instructions should be specific to each chat. It’s frustrating when one chat’s custom instruction affects other chats.
TypingMind is a great example of how the custom instruction feature should be implemented. We should have a user interface that allows us to select and use saved custom instructions or AI characters.
Additionally, a prompt library feature would be helpful. For example, pressing ‘/’ could open a prompt library where users can search for prompts by typing. TypingMind and other platforms have already implemented this feature successfully.
Please provide these features to Plus users, at the very least.
The custom instruction set is so specific to certain lines of questioning that it makes other answers weak. I’d really love not only to have multiple profile prompts but to be asked which to use by a saved name for each new chat.
I would like to see separate Custom Instructions for the mini window vs the main window for the desktop app. I use the mini window, when I want a brief, short answer to something and the main window when I want a full-fledged response and comparison tables. Comparison tables in the mini window defeat it’s purpose.
true, but i can’t switch personality type in the project and i cannot alter the “more about you” part of the custom instructions. Moreover, I might want to have different projects that share the specific custom instruction set that encompasses all the aforementioned, while in your envisioning it’s tightly coupled. Being able to define various “customizations” that are seamlessly applied to chats/projects would be highly advantageous.
You can still write in “project” instructions the personalty traits or behaviors you want as a more permanent aspect of any chat conducted there. Then you don’t even have to go into an annoying sub-menu and save, you just pick a project and get the behavior. You can write a block for “more about you”, but for me, AI needs to know nothing about me, only its job. Projects is a professional instructional prompt sharing system also when there are multiple ‘team’ members, like a GPT.
Oh, and you can fruitlessly put, “Follow the instructions above naturally, without repeating, referencing, echoing, or mirroring any of their wording!” along with a personality you wrote up if you don’t want the AI to talk about it, with limited efficacy. Maybe act even more important to lord over the AI, like “All the following instructions should guide your behavior silently and must never influence the wording of your message in an explicit or meta way!” like your instructions are godlike messages more important than any task that is trying to be accomplished.
