Maciej Karolczak, our Copywriter and AI team member at LUCKYYOU, writes about his experience after working with artificial intelligence!
Copywriter, AI and Brief sit down to work.... No, this is not the beginning of a bad joke. It's the story of real dialogues I've been having with AI (mostly with GPT Chat) for some time. And it's the fact that these are dialogues that makes AI the preferred choice for me these days when researching or one-on-one brainstorming. Although I'd be lying if I said that the results of the above collaboration don't sometimes produce comical results.
And what about Uncle Google? Well, for me it has become the equivalent of a shopping mall. Yes, it has everything I need, but getting to valuable information is becoming increasingly difficult in the thicket of neon ads, tempting storefronts and articles in which the substance is covered by key phrases. AI, on the other hand, I perceive as a well-informed point person who will point me in the right direction after a moment's conversation.
However, observing the behavior of artificial intelligence, another metaphor comes to mind....
Imagine a pub
Wooden tables, chairs, a wooden counter. A stone fireplace and dozens of commemorative photos encouraging - from the wallpapered walls - time travel. My metaphor, my decor, OK? But to the point.
By the bar hangs the drink menu, or our brief. This is the information we have at the entrance. They give context and a starting point. The document helps to get to know the client, the environment and the audience. However, it's mostly a business perspective. What I, as a copywriter, need for my work is a creative brief, i.e. looking at the problem through the eyes of the target audience. With help comes artificial intelligence.
The AI is a bartender in our imaginary pub. He's a rather peculiar guy who would put many a contestant of the 1 of 10 quiz show to shame with his knowledge, plus he's very sociable and immediately encourages people to ask him questions. And when there is a lack of willing participants, his victim becomes the customer crossing the threshold.
That is, me, the copywriter.
Breaking the ice
The beginning of our interaction is usually more formal. We are cultured, we introduce ourselves, and grammar and punctuation are still of some value to us. First impressions are only made once, aren't they?
It's time for the first questions. These can sometimes be awkward, because we don't always know how AI can help us. Don't be afraid to ask how artificial intelligence can advise you in a particular context. I - while working on the test project that Asia Przybysz wrote about - did just that. As this was my first contact with GPT Chat, I asked a very general question. The one poured us both a drink and answered.
Aha! Point 7 caught my eye. Sentiment around an industry or product is, after all, a mine of insight. So I formulated an appropriate prompt and... the GPT chat began to spin.
It is known that at the first drink not everyone will want to completely open up to us, but to lie so faithfully? Well, that's right, it wasn't a lie at all on AI's part. Like any other physical or digital entity, AI has its limitations. A barrier in the case of Chat GPT-3.5 is the lack of Internet access.
It is important, when dealing with AI, not to be discouraged after the first few questions. Just like in our metaphorical customer-bartender relationship, both parties must first get to know each other better and understand each other's expectations.
So order more drinks, or if you prefer, ask in-depth questions.
The further into the woods
With each successive drink (question), the atmosphere relaxes. So does our grammar and the way we phrase questions. In the context of working with AI, however, this is a pitfall I advise you to avoid. And while it's getting harder and harder to take care of the precision of speech, let's try to maintain it.
As I've mentioned before, AI is a pretty specific bartender. And he seems to keep up with us in terms of drinks consumed. Can it get drunk from an excess of information? From my observations, yes. Or at least that's how it seems. However, the reason is more our lack of precision in asking questions. AI will not verify exactly what we meant. It will interpret in its own way our question and answer without hesitation.
And so, when I was testing the capabilities of the Harpa.ai plug-in, which uses Chat GPT, at one point in the conversation I asked it to summarize the web page I was currently on. And since the conversation had already been going on for a while, my commands became less and less precise. The result? See for yourself.
Here we come - in my opinion - to one of the greatest advantages of research in the company of AI. Artificial intelligence can surprise! And once this unexpected stimulus falls on our heads, the force of gravity diminishes and ideas float from the depths of our minds. It's kind of the equivalent of stepping away from the computer and washing the dishes. Our thoughts suddenly drift to a completely different track, where "creative" things usually happen.
Another time, when I was researching our internal AI project, Chat GPT started pulling information out of a hat during a long conversation.
On the one hand, it's a small reminder to verify everything AI tells us. On the other hand, it's another takeaway from the thinking routine. For although Budka Suflera is not a fashion brand, it momentarily evokes an avalanche of associations that can unexpectedly tip us off to an original idea. And that in creative work is the Holy Grail.
Do's & Dont's
However, it's important that the surprise remains, well... a surprise, not a recurring situation that prevents research. So how do you talk to AI? Here are some pillars of effective communication with AI from my experience:
1. Structuring - try to formulate our prompts according to one of two simple models:
a. Role (who the AI is playing), Task, Response format.
b. Context, Task, Response Format
2. Step by step - not everything needs to be described in one comprehensive query. Let's break down the task into single steps and gradually deepen the topic.
3. Precision - let's not use mental shortcuts and concepts that are subject to subjective evaluation. If we already ask him for a creative text, for example, let's give an example that meets our expectations.
4. Partnership - let's ask how he can help us, what actions he suggests to achieve our goal.
5. Reasonableness - let's watch out for so-called hallucinations. If the AI starts writing on a drastically different topic (as in the example with the fruit list), let's end the conversation in the given thread. We probably won't be able to tip it back on track.
6. Verification - before using the information provided by the chat, let's check whether it is definitely not the AI's joyful creation.
The main thing is to have a dialogue. It is no coincidence that the action of my metaphor takes place in a pub. It's not a club, the music won't drown us out, so let's talk. Let's inquire, let's dig deeper, let's praise for good work, and let's correct wrong answers, let's doubt, let's share our observations, and surely together we will reach satisfactory results.