Is it actually humans being tested when using AI? ~On vocabulary and structural skills.~

Hello, I'm Takiguchi, in charge of marketing at Blastmail.

Recently, the opportunities to use Gemini and ChatGPT in my work have increased dramatically.

As I wrote in a previous article, even someone like me with zero programming knowledge was able to have Gemini create a "diagnostic tool" in 30 minutes, and I am constantly amazed by the evolution of AI.

However, as I face AI every day, there are moments when I suddenly feel a chill. When AI gives me an amazing answer, I think, "Wow, that's smart!", but conversely, when it returns a nonsensical answer.

In the past, I might have laughed it off, thinking, "Well, AI still has a long way to go." But now, I think this.

"Wait, wasn't this just because my instructions (prompts) were bad?"

The more AI evolves, the more "the level of the human user" is explicitly visualized. Today, I would like to write about "vocabulary" and "structural skills" that are being put to the test in this AI era, with a note of self-admonition.

Actually, I have a little bit of confidence in my vocabulary.

I worked at a publishing company in my previous job and was in charge of online sales such as Amazon, so I used to think about copy for the site and banners for site placement.

Although the designs themselves were created by designers, I always made sure to specify the tone and manner using multiple words when conveying the image.

Also, since there were many editors around me who faced words every day, I think I had more opportunities to face "words" than the average business person.

So recently, it's not exactly "relying on old skills," but I'm happy that my experience from back then is being used in the form of prompts.

Prompts are a mirror that reflects the "resolution of your own thinking"

When using AI, I am made aware of the cruel fact that "the content of the prompt (instruction text) depends entirely on the user's vocabulary and structural skills."

For example, it was the same when I had Gemini create a diagnostic tool.

When I first just threw in "Make a diagnostic tool like MBTI," it returned something that looked the part but was empty inside.

AI is not to blame. It just interpreted the vague instruction "like MBTI" vaguely.

So, I changed the instructions as follows.

1. Objective: I want to create a diagnostic tool for choosing an email delivery system

2. Options: I want you to differentiate between Blastmail, Haihai Mail, etc., based on these conditions

3. Output format: I want it to be code that runs in a browser using HTML and JS.

Only after verbalizing (structuring) it to this extent did the AI provide a "usable deliverable."

In other words, how accurately can we convert the "finished image" in our heads into "words"?

The mirror that is AI reflects our lack of vocabulary and shallowness of thought exactly as they are.

Vocabulary for conveying tone and manner in prompts when generating banners

There may be some who are surprised, but I have Gemini's nanoBanana create most of the banners for the articles posted on this note.

Here, I will show you the prompts used for the last three banner generations.

I asked for this quite simply and it came out looking good, so I just used it as is. This isn't very helpful, is it? Gemini is amazing.

The additional prompts I used were "severe" and "serious"
...Those don't seem like phrases typically used for tone and manner...?

This is for an article I haven't published yet.

When I specify "POP," it usually becomes incredibly colorful and looks like jelly beans, so I made sure that didn't happen.

And for this article.

"Business-like," "Blue-based."

Having shown you this much, I will introduce two of my own personal styles for tone and manner.

Don't get too hung up on adjectives

Since the purpose of tone and manner words is to "align images," I don't think you necessarily have to stick to adjectives like "bright," "fast," or "cool."

"Severe" and "serious" are adjectival verbs, after all.
I also sometimes use sampling like "something like XX."

Fancy words are unnecessary

Instead of overthinking, it is important to keep a variety of simple and easy-to-understand expressions in your mental drawer.

While difficult words leave little room for interpretation, they also carry a high risk of not being understood at all.

Rather than trying to hit a pinpoint target, the idea is to use multiple broader terms to fill in the Venn diagram.

AI is a 'highly capable subordinate who just doesn't read between the lines'

Well, this discussion—or rather, this content—has been talked about since the dawn of ChatGPT about three years ago, so there might be no need to write about it in detail.

Thinking about it this way, the skill of using AI could be said to be 'management ability' itself.

AI is a super-elite subordinate who never complains, never gets tired, and possesses vast knowledge. However, it has one flaw. It 'doesn't read the room (doesn't read between the lines).'

If a boss (me) gives a vague instruction like, 'Summarize the materials in a nice way,' the AI won't know the definition of 'nice' and will just create a document filled with generalities.

On the other hand, people who are good at using AI give specific and logical instructions.

'The target audience is [segment], and the goal is [objective]. Summarize it in 3,000 characters following the [framework] structure. Use a business-like tone.'

This is exactly the same as 'giving instructions as a competent boss.'

The theory that 'people who can use AI' have the makings of a 'good boss'

People often say that 'AI will take our jobs,' but to master AI, you need 'advanced instruction skills (direction)' that AI itself lacks.

• The ability to set goals (conceptualization skills)

• The ability to break down tasks and communicate them (structural skills)

• The ability to verbalize vague images (vocabulary skills)

• The ability to judge the quality of the output produced (evaluation skills)

These are all skill sets required when humans manage subordinates.

Thinking about it this way, perhaps those who can use AI well have the potential to be 'good bosses' even when dealing with humans.

Conversely, those who get frustrated with AI, thinking it's "useless," might also be getting frustrated with their subordinates, thinking, "Why don't you understand without me having to say it!" (The boomerang effect hurts...).

It is always the humans who are being tested.

AI is not a magic wand. It is an "amplifier" that magnifies the abilities of the human using it.

If the thoughts of the person inputting are organized, AI will respond with 100 times the speed and quality. However, if the thoughts of the person inputting are full of noise, AI will simply return that noise magnified 100 times.

"Using AI" is like a form of training that forces you to confront your own "ability to articulate" and "logical thinking skills."

The more technology evolves, the more it seems that what is ultimately questioned in the end is our human "fundamental intelligence (the power to put things into words)."

I, too, want to strive to ensure that Gemini doesn't get exasperated with me, thinking, "This boss's instructions are sloppy..."

Request for information exchange regarding the newsletter

Thank you for reading the article this far.

This time, the topic was unrelated to email marketing, but I would be happy to exchange information regarding these kinds of "AI utilization" and "management" concerns.

Whether it's "our AI utilization techniques" or "worries as a boss," or even sales, anything is fine.

I am just an ordinary marketer, so let's have some casual chats about real-world situations.

Schedule a meeting here>>
*This is an online meeting with me.

Free trial available

If you get tired of giving instructions to AI, why not try delivering your own words directly using a simple email distribution system once in a while?

(That suggestion is a bit of a stretch, isn't it.)

But if you are interested, please feel free to try it out.

Sales partners wanted

For those who have client work or distribution agency projects, and for all affiliates, Blastmail has a sales partner program with a stock-type revenue model.

If you are interested, please apply. (Registration is free)

Click here for case studies