Do you remember your first experience with ChatGPT (my metonym for LLMs)?
I do, and I recall a shiver — entering a sort of open-ended stillness. A realisation that everything is now different.
Many feel a disconnect between an internal idea and the ability to express that idea. I do, resulting in feeling stupid and inadequate.
I struggle to express myself, especially with the written word, stumbling and mumbling. Closely linked to this is trusting and validating the authenticity of my idea. In the face of much authentic knowledge, expressing a thought in unfamiliar territory is terrifying.
Enter centre stage, ‘Chatomir of Deeplore’ from the land of ‘Aithrendor’. This great wizard, with a small incantation, can:
- Generate content of great scope;
- Present the most impeccable research.
Beware the Sirens of Homer’s The Odyssey. That their sweet song devours your authenticity and credibility ends on the rocks.
Using LLM’s as a research assistant is common practice now. Trusting the output is a different issue. LLMs are brilliant at fabricating seemingly credible answers.
We are flooded with AI-generated content. I experience it as devoid of human connection.
Yet in these winds, It is imperative to understand generative AI. For our sea chest, it is an essential tool. Without it, we will flounder. Of this, I am certain.
My experimentation and research led me to a technique which moves my compass closer to trusting an output. That is the use of ‘Prompt Commands’.
I use these commands to ‘interrogate’ the LLM output. To get greater detail and in-depth context. This allows me to discern. To accept, reject or iterate further.
Even after this, I am still working with a hypothesis that is tested in a real-world scenario.
The prerequisites for this tool to work are:
- Be clear on your objective;
- Remain connected to your authenticity;
- Retain your authority — do not abdicate to the LLM;
- Be responsible and accountable for how the output is used;
In [my] lay terms, a prompt command is an instruction linked to a process to achieve a specific goal. The command is embedded in your prompt.
- Command;
- Purpose;
- Procedure;
- Example Use.
A command can take inputs or parameters.
Create commands specific to an outcome or task. Such as:
- Get more detail;
- Get citations;
- Analyse;
- Compare;
- Do a specific web search.
Here are two examples.
## Expand Detail
Command: /expandDetail
Purpose:
To enrich the conversation by providing expanded information, deeper insights, and comprehensive background, thus enhancing understanding of the topic discussed.
Procedure:
Receive a topic or statement from the user that requires expansion. Use knowledge databases and model's training to gather comprehensive information related to the topic. Include detailed background information, relevant contexts, and thoughtful analysis to enrich the user's understanding. Present this expanded content in a clear and structured format.
Parameters:
'topic: string' (the main subject to expand on), 'depth: string' (desired level of detail: 'low', 'medium', 'high').
Example Usage:
/expandDetail topic="paste an extract of output you want additional detail on, here" depth="high"
## Cite Detail
Command: /citeDetail
Purpose:
Cite the sources of information used in the GPT's responses to ensure credibility and traceability.
Procedure:
Identify the part of the conversation or the specific information that needs sourcing.
Retrieve the sources from the model’s training data or simulate potential sources based on the data range and type.
List all relevant sources and their detailed descriptions that back up the information provided.
Ensure transparency by explaining how these sources contribute to the credibility of the information.
Parameters:
'details: string' (the details or facts to cite), 'format: string' (citation format: 'APA', 'MLA', 'Chicago').
Example Usage: /citeDetail details="Your last output" format="APA"
So:
- Craft your prompt in the manner you would normally;
- Add a section to you prompt with the heading ‘# Commands’;
- Paste the commands;
- Use the command as indicated by Example Usage
To see a full prompt using commands, jump to the Completed Prompt in this article. >>>
These two useful GPTs will help you create your own commands.
Command GPT >>>
GPT "Prompt Command" Maker >>>
Use this prompt to generate the commands.
I give you a list of commands to ‘/create’. Please step through the list and ‘/create’ the commands in Natural Language format. The commands specifically relate to and are applied to the LLM output. The commands are used in the conversation flow to enhance the feedback loop within an existing conversation. Commands should contain attributes ‘Command’, ‘Purpose’, ‘Description’, and ‘Example Usage’. Add the additional attribute 'Parameters' if required to improve performance. The Command should begin with '/'.
--------------------------------------
- Validate Output - Give a detailed justification for the output given. Motivate your choices and the reasons and logic behind the choice.
- Expand Detail - Tell us more about this, adding lots of details, background, and your thoughts to help us understand better.
- Cite Detail - Tell us where you got your information from, including all the details.
---
If you find this content useful, please share it with this link: https://patrickmichael.co.za/subscribe >>>
Take me up on a free 45-min prompt workshop: https://patrickmichael.co.za/form/gpt-workshop-leadgen >>>
Are you looking for a GPT to conduct research? Start here: https://patrickmichael.co.za/power-your-solopreneur-research-essential-gpt >>>;
Until next time,
Patrick