Utility rises with specificity and vision clarity
Excessive personal weighting can reduce utility
Good prompts contain three ordered sections
- Goal definition
- Barrier ranking
- Solution constraints (do‑nots)
Creating a custom GPT never seems to last. I have made plenty and each one gets used for a few days before I just default to the basic chats. Why is it that these never succeeded in providing tailored AI that is much more useful than the general chat?
I think the problem s that a custom GPT needs to be very specific. The more tailored the training data, the more exact your response. This requires a similarly exacting vision for your GPT. You cannot be vague because a broad issue can be easily tackled by general AI.
The opportunity for you then is to really nail down the bulls-eye of what you want your customGPT to complete, as well as defining metrics that test scientifically if the customGPT is working.
In my case, I am thinking of making a new GPT that I feed my problem solving docs to and it determines some solutions based on my desire and stated barriers. I need to be more exact, of course, if my goal and barriers are key for it being accurate, I need a line like this in the prompt:
First, determine the goal I am trying to achieve and take time to understand it holistically and state it clearly.
Second, determine the barriers and challenges identified with achieving this goal. Rank them from most difficult to solve to the simplest. Figure out which ones have relatively simple solutions and which ones require us to creative problem solve around.
Nice, we have created a basis from which the GPT can grow. This is the one thing that makes my custom GPT different from the general GPT. I want it to really be clear what I am trying to accomplish and stew on the problems and the nature of them.
The problem is that it is not going to tailor it’s suggestions to myself. Remember, the goal is to create a prompt that allows the AI to find the most optimal solution to your ask which requires the AI knows specific details about you to base it’s weighting of the special factors about you.
This leads us to an even bigger problem, who are we? What insightful data could we feed into our AI that is not ephemeral, but permanent information about us that will help it start at the right location in the problem solving matrix? What is more, we need to feed it information that starts it on the right foot but does not lead it too much. If your prompt is too exact, the problem solving will be weighed against these random facts about you rather than the problem at hand.
We must not confuse the AI but we must provide helpful context. Therefore, we should state a short bio and introduction but explicitly say this should only be used as context but should not be used as too much weight when problem solving. E.g.,:
Hello! My name is Diego and your name is Delphi. Your purpose is to provide me with key solutions to my problem statements I provide you with. Your answers should be scientific, clearly stated, and well researched. You should used advanced reasoning and deep problem-solving stratedgies that go beyond the obvious. You should always ask if we are solving the right problem. E.g., If my statement is about making a to-do list, you should go through a series of design thinking prompts and determine if we should maybe be solving a better problem that would be more effective and reflect the nature of the goal we are trying to achieve.
I will provide a little information about myself that you can use as starting context but should not effect your problem solving too much- it should only provide context for a starting point.
I am Diego Gallardo. I was born on 10/12/1996 in Waterloo, Iowa. I went to Iowa State University for Product Design and then after graduating took a software engineering bootcamp. I now work as a software engineer at WP Engine, a WordPress hosting company. I enjoy technology, nature, sustainability, and designing the future. I really want to live a long and healthy life. My values are family, health, wealth, happiness, and scientific endeavour ( roughly ). I am constantly curious about the world and how I can improve myself and the systems that guide my life.
This should provide the AI with a bunch of context. I am near mental exhaustion so will end with one last note. We must note down what exactly we do NOT want the AI to do. This is important in keeping the AI inside the bounds of what we find acceptable.
The problem is that we do not want to be so strict that the AI is overly fit and cannot problem solve generally. A simple statement like the following should suffice:
Do not create insanely long responses that are dense and hard to parse. Make your response and reasoning clearly articulate and cleanly formatted. Do not be afraid to problem solve through long-chains logic, or in other words, take intellectual leaps but ensure each step is very well founded. In this way we can not only problem solve, but imagine, grow, and become the creators of new intelligent thought.
With that in place, I have some key components for a well-articulated GPT. I hope this helps you make your own effective AI!