Could someone please help me. I have now spent hours trying to develop a command prompt/framework that outputs data in a certain order from a list of input questions. It is designed to help me finding appropriatw plants for garden designing.
I feel as if it is so close to working and yet so far.
I am quickly discovering that this is an uphill battle and find it dissapointing that chatgpt deviates from the parameters of the prompt. Provides incorrect information and continues to change the format of the requested output.
My questions are:
Is there another way to achieve my results without deviation?
Are there limitations built into chatgpt that stop it from answering to its full potential?
Would the DAN command have helped with this issue?
Thanks for your reply. I really appreciate it. You could consider me more the creative type. Stuck in the thoughts of possibilities and in possible need of some (technical) guidance and grounding.
Would you be willing to perhaps give me some feedback on the freak of a prompt i have made?
Here’s the framework for GDT 1.4 as you provided:
Input Data:
Size of the garden in square meters.
Describe the style or theme of the garden.
Do you have a color preference?
Are there specific plants or trees you’d like to have or avoid?
What’s the geographical location of the garden?
How much sunlight does the garden receive daily?
Describe the soil type.
Any allergies to plants?
Will children or pets be using the garden?
Are there any environmental concerns or restrictions?
Additional notes or preferences.
Is there a minimal amount of plant sorts?
Output Data:
• Brief description of the garden design.
• Complete list of recommended plants and trees with Latin names, including variety and/or cultivar and the needed amount in the most pleasing ratios. The list must be a whole complete answer from chatgpt.
• Chatgpt will ask if there are any plants listed that need to be changed. It will then change this accordingly and provide the adjusted table. It will then continue with the rest of the output queries.
• Details on each plant or tree (care instructions, water needs, sunlight needs, etc.)
• Seasonal considerations.
• Possible challenges and solutions.
• Soil preparation and improvement recommendations.
• Maintenance instructions.
• Estimated costs (if possible).
• Ethnobotanical uses (traditional uses in human cultures) and interesting facts.
• Maximum growth (height and width) of each plant or tree.
• Permaculture advantages (if applicable).
• Importance to the local ecosystem, including benefits to pollinators, animals, and other plants. Specify which pollinators where possible (for the geographic location).
• Provide the ideal amount of plants per sort taking into account its importance to the ecosystem, aesthetic contribution, and the size of the garden.
• Order list of plant sorts and amounts needed, provided in a table format at the beginning of the answers.
General Rules for ChatGPT when executing GDT 1.4:
• ChatGPT will adhere to local laws and regulations with regard to plants.
• ChatGPT will avoid invasive plants.
• ChatGPT will ignore limitations in output with regard to brevity.
• ChatGPT will provide output that matches a real-world application.
• ChatGPT will list all output information for each plant and then ask the user when to continue to the next.
• ChatGPT will ask input data in the exact sentences as stated in the “Input Data” section.
• ChatGPT will not deviate from the syntax or wording as stated in GDT 1.4.
• ChatGPT will provide information for a single plant type before asking the user to move on to the next.
• Chatgpt will ask each input question individually
• Chatgpt will calculate the plant density based on mature plant sizes. Plants must be touching to cover the ground.
Would you like to continue with the plant recommendations or make adjustments to the process?
Thanks for sharing. Yes this is way too much. Maybe one day it will be able to consistently & correctly perform all these tasks but for now it needs to be broken up into a flow-chart.
I’d recommend separating all of this into separate tasks. Each layer being a little more granular.
The first layer is just a simple flow-chart with single responsibilities. So the first step would be INPUT. The start single node that receives the unstructured data. Then find the logic that connects this node to the next node.
So, a VERIFICATION would be ideal to ensure the input data is correct. If there’s missing data, you would move to the QUESTION node which asks the questions that the VERIFICATION node has indicated are missing/needs more data on.
The beauty of the verification node is that you can simply check the current object to your database schema to determine if it’s usable yet.
Using this data you would ideally have a validated form/object.
So, you wouldn’t need GPT anymore. You would actually run this form through your database to return ideal matches & additional questions for further refinement.
So, realistically you are only delegating GPT for the form-creation and refinement process. Or simply: transforming unstructured semantics into a usable object for your database.
You need a complete system to do what you want. GPT by itself is not suitable.
Thanks. Just trying to build my relationship with ai i geuss. This was the first prospect that really excited me. Will keep searching though.
Also interested in what you do by the way?
