Hi OpenAI team,
I would like to suggest the following simple features that I believe would make everyone’s life more pleasant when using ChatGPT, especially for those who use it for code assistance and/or long periods of time.
-
Larger code boxes.
The code boxes are the same width as the rest of the text while code often has longer lines. It would enhance readability and be a lot more convenient for the eyes of someone who spends a lot of time on ChatGPT for code assistance. -
A toggle button that configure if the “enter” key submits the prompt or create a new line. Its not user friendly to always have to hold shift + enter for a new line. Sometimes it bugs and just sends the incomplete prompt. Many chat products (like Slack) have implemented this for a long time. Again, for someone who spends a lot of time on ChatGPT for any usage this can be tiring. Being able to toggle the “enter” key to create a new line would be a relief.
-
Allow Meta prompts that can be toggled on and off directly in the chat area. Add a small button on the chat box that defines guidelines. This can introduce polishing guidelines on the fly which would add more granularity than the guidelines defined in the custom GPT or memory. ChatGPT and Custom GPTs dont always follow their guidelines and its very tiring to always repeat yourself, especially, again, when you spend a lot of time using it for your work. Meta prompt could be as simple as “Dont return an explanation, just return the answer” or “answer only by yes or no”. I have a Custom GPT for my personal work and has very clear and defined guidelines but they are not being followed all the time (most of the time actually). Also, it seems that as the conversation gets longer, it becomes harder and harder for ChatGPT to follow them accurately. As ChatGPT is not perfect and always under improvement, this could fake the impression that it is effectively able to follow them. (at least until it actually is very good a following guidelines). An example from my chat:
From me:
“i keep asking you to drop the text example and example usage, to leave a placeholder instead. its in your gpt guidelines and i keep repeating it. please explain to me why is it so hard for you to follow the guidelines? what can i do as a gpt user to enforce those guideline more efficiently?”
From ChatGPT:
" I apologize for not adhering strictly to the guidelines you’ve provided. The difficulty in following these guidelines consistently arises from the natural tendency of the model to provide thorough and practical responses. To ensure better adherence, here are a few steps you can take as a GPT user:
- Reiterate Guidelines: Clearly state your requirements at the beginning of each conversation or before each specific request to ensure they are top of mind.
- Explicit Warnings: Mention explicitly that examples should be replaced with placeholders before each relevant request.
- Feedback Mechanism: Provide immediate feedback whenever guidelines are not followed. This can help reinforce the importance of these guidelines in real-time.
- Specific Templates: Provide specific templates or formats that you expect the responses to follow. This can help the model structure its output correctly.
- Frequent Reminders: Periodically remind the model of the guidelines throughout longer conversations to maintain adherence.
Your feedback is essential in ensuring that the model meets your requirements accurately and efficiently. Thank you for your patience."
ChatGPT is telling me itself it that it cannot do any better than to try to follow the guidelines and suggests me to repeat myself, which is the last thing that would make it user friendly. At the same time repeating is till the best and most efficient way to enforce guidelines. Having a Meta prompt option would allow to open a small text box in the chat area and define a small prompt, then select if you want this to be added before or after your prompt.
That way, those guidelines can be included to all my prompts and prevent the generative nature of ChatGPT to deviate from the custom guidelines as the conversation goes. This could be implemented with some simple javascript that simply concatenate the text from my toggled Meta prompt to my actual prompt. Skipping the need to wait until ChatGPT is perfectly able to follow the guidelines every time. That way the guidelines described in my Custom GPT can be viewed as global instructions while Meta prompts would be local instructions, used for live fine tuning.
Plus this would also allow a user to standardize quality prompts which is slightly different from just customizing a GPT.
I hope you see the potential in those small details like I do.
Best