There has been a flood of material published lately on the many varied uses that can be made of AI tools like ChatGPT, Chatsonic, YouChat, and MidJourney. The more helpful material, I find, focuses on the intelligent framing of prompts to engineer optimal responses from the AI tool.
The notion of prompt ‘engineering’ is not new of course. Neither is the use of a class of algorithms called ‘mental models’, which can help with decision-making, information processing, and problem-solving. Likewise, mnemonics help people to remember certain patterns, or offer template outlines (i.e. prompts) of the parts of a system or process. Some of these have been used for generations. Both AI and People prompts are addressed in this post.
AI chatbot prompts
Much like a recipe, or a mathematical formula, the quality of prompts used in AI chatbot interaction plays a significant role in determining the quality and useability of the outcome (response). While these tools use Natural Language Processing, they ‘prefer’ to read the prompts provided in a form somewhat like the code a programmer would use at a Command Line Interface.
A Command Line Interface (CLI) is a text-based interface for interacting with the computer, where the user types in code (commands) to perform actions. On the other hand, an AI chatbot prompt is a conversational interface where the user can input natural language queries and receive text-based responses from an AI language model e.g., OpenAI’s GPT-3. In essence, a CLI is more structured, and a chatbot prompt is more flexible and user-friendly.
ChatGPT provided the following response to my query about how it processes a prompt:
Me: How does the natural language processor used by ChatGPT process the prompt entered by a user?
ChatGPT: The natural language processing (NLP) used by ChatGPT is based on the transformer architecture, a type of deep neural network. When a prompt is entered by the user, it is tokenized, that is, broken down into individual words or subwords, and these tokens are then passed through the NLP model. The model has been trained on a massive corpus of text data, so it has learned patterns and relationships between words and phrases. Using this knowledge, it generates a response to the prompt based on the context of the input, as well as its training data. The response is then generated by predicting the next token in the sequence given the context of the input and selected from a set of possible outputs. The final response is then de-tokenized and presented to the user as a natural language text output.
Various online sources offer advice or instructions on how to engineer prompts to make more effective use of AI chatbot tools. I thought I should ask ChatGPT what qualities it recommended in the framing of a prompt. The result is summarised in the chart below.
Prompts for people
Previous posts have highlighted several mnemonics and mental models to aid strategic planning, governance, project management, etc. STEEPLE analysis for evaluating developments in your external environment, and the SOSTAC framework for internal situation analysis and response, are just two examples.
Every field of human endeavour has its own large glossary of acronyms and mnemonic terms, which are used as shorthand for longer and more complex names or descriptions. For this post, let’s look at just one human activity – writing.
Writing is used in most fields of work, and there are many prompts available to help people who wish to write for different purposes. A selection of these is illustrated in the chart below.
While the chatbot prompt is the human half of a conversation with an AI system, writing prompts are more like frameworks (or section heading outlines) that you can adapt to suit your writing needs.
Your board paper template
In my previous post (Prioritisation of everything, everywhere, all the time), I noted that placing your recommendation (and/or purpose) at the start of your board paper was a helpful way of focussing directors’ attention on the outcome proposed.
Two board paper prompts (mentioned in the chart above) are illustrated below, and while their acronyms are not memorable, each offers benefits worth considering. If you don’t already have a standard template, you may find it helpful to adapt one of these for your purposes, especially as it will assist your directors to use their time more efficiently, and to focus on the key decisions to be made.
See also:
Continuous Reflective Governance
