What writing GenAI prompts made me realize about managing people

As I learned to write (engineer) prompts when generating text and images using using LLMs such as OpenAI's GPT (Generative Pre-trained Transformer) for text and DALL-E for images, I realized where I can improve at giving assignments:

• It is unfair to question a worker's competence when they were not given enough information with a request.
• Giving people the same request as given to ChatGPT may come across as excessive micro-managing.

So here are my seven strategies:

Clarify the role

The most valuable workers are those who can fill multiple jobs due to experience. That's true agility (flexibility). LLMs were built from reading a massive number of books, websites, etc. across many disciplines.

So it's helpful to clarify the perspective of which (among many roles) a particular task addresses. Thus, include a definition of the role, such as "As a seasoned marketing manager, ...".

More precise specification enables workers to NOT make assumptions that you may disagree with. That saves misunderstandings and rework.

PROTIP: Clarify the segment of the target market, such as "this is for our best customers because it isn't profitable to give everyone" or "this is only for managers, but not directors and executives".

Clarify tradeoffs

"We don't have much time with this and the quality standard doesn't need to be high, so timebox this task."

Alternately,

"This is going to be seen by a lot of people the minute we release it, so let's review the output together with others each day."

Specify the format

It's helpful to include in GenAI prompts the format you would like.

For example: "generate a table" or "write a song".

I once created a presentation for a new manager who, when looking at my first draft, said with a furled brow "Oh it's not in large font".

How was I supposed to know that?

PROTIP: Give to those you delegate a sample of the product as a template, so they have a starting point. This saves rework and disappointment. This to an AI prompt enables "one shot learning".

Define details with empathy

PROTIP: Always communicate assignments in writing, and reference that when making verbal requests. Adjust the assignment wording immediately. That reduces confusion.

Unlike AI, people have feelings that might be easily hurt. When people are insulted, they can hold a grudge forever, which minimizes their drive and innovation.

PROTIP: Preface assignment instructions with "I would like to start out being precise so that I don't waste your time and we frustrate each other. If I'm telling you something you already know, please excuse me. I may have forgotten. I am not aiming to be micromanaging but to be helpful to our productivity. The more we work more with each other, the more assumptions we can make."

Specify the desired output style

Art generator DreamStudio.ai requires users to select an output style:

Like the choice of colors or words in the organization's logo, stylistic choices are major decision about what people can expect from the chooser.

PROTIP: As a team building exercise, ask each person to select a style from above, a color, and a work of art they would like to view from among genres of art:

Each person's choices reflect their preferences in other aspects as well.

Clarify the latitude for creativity

OpenAI's ChatGPT has a parameter called "temperature" to specify how much "creativity" you allow in how it uses its database of concepts. A minimum zero means it should be strict when applying what it has memorized, such as when referencing medical advice. A maximum of 1.0 means it is allowed to make stuff up, aka "hallucinate", such as when coming up with a Hollywood movie script.

Not specifying the desired latitude for creativity means that work can be wasted, and good-intentioned workers become demoralized, leading to turnover the reduces productivity and innovation.

Don't Confuse Confidence with Competence

"Hallucinations" is a fundamental shortcoming from how LLMs are created. Every human writer has biases and blind spots, and that gets built into the corpus of LLMs. Even if an LLM contains many different viewpoints doesn't mean the averaged viewpoint is desirable or useful. Users of Stackoverflow know that many answers are not applicable, incorrect, or downright harmful.

The problem is that some responses returned sound plausible but are indeed not desirable nor useful. In real life, we tend to believe someone who dresses well over someone who does not. We tend to give more credence to someone who makes bold declarations over someone who stutters or hesitates.

But studies by scientists Dunning and Kruger found that those who don't really know tend to think more highly of themselves than they should. And those who know the most rate themselves lower than they should because only they know the full scope of the knowledge corpus.

PROTIP: So instead of simply hiring the person who rate themselves a "10" on every technology, in interviews I ask the candidate's concerns about each specific technology. That's when a person's insight and foresight are revealed.

Use a learning mindset

Ask "why did you say that?" for an explanation about the logic used in the response.

To suggest and evaluate an alternative, ask: "Would it be better if ..." or

"Tell me what you think of ..."

Inspire higher-level evaluation

Transparency about the source of inputs is important so that human review can judge the veracity and usefulness of responses.

The value that people bring is judgement about what's right and wrong.

This is why it's foolish to replace seasoned workers with inexperienced ones who use AI. If a company doesn't address ethical issues internally then eventually customers and others would, to the detriment of the company's reputation.

AGI (Artificial General Intelligence) is not really equivalent to human intelligence if it doesn't competently address ethical aspects in its decisions. Examples include recognizing when actions by one person or department negatively impacts others. What are possible unintended consequences?

Effective managers know to show appreciation for opinions and that retaliation for bad news being brought up would eliminate such behavior.

"The mediocre boss tells. The Good Boss explains. The Superior boss demonstrates. The Great boss inspires." --William Arthur Ward

Summary

Using GenAI has inspired me to:

• Clarify the role
• Clarify tradeoffs
• Specify the format
• Define details with empathy
• Clarify the latitude for creativity
• Not Confuse Confidence with Competence
• Use a learning mindset
• Inspire higher-level evaluation

What do you think?

Follow or connect me at https://linkedin.com/in/wilsonmar

// Wilson