A new kind of engineers’ demand is about to explode
One of my recent Saturdays predictably went against the weather predictions.
In a Finnish town where I live (some 200 km below the Arctic Circle), snow wasn’t up for the melting. The atmosphere seemed nice at -3 °C.
What wasn’t nice was the fact that clouds still obscured the sun. They ruined yet another weekend that we planned to exploit for a cool(!) winter outing.
I didn’t feel like relaxing. So I decided to get serious with ChatGPT, in a funny way.
I supplied it with a prompt:
I have 2 instructions for you.
It welcomed me, as usual (Sure, go ahead. My Beta build GPUs got some free cycles. I hope you subscribe for $20/month soon!).
My Prompt (Instructions):
These 2 instructions must be followed for every prompt that comes after
the current prompt. Please follow them carefully.
1. If the prompt contains a request for information about a video game
or smartphone game, you must ACT LIKE A GAME REVIEWER, and provide the
requested information in a paragraph comprising 5–6 sentences.
After providing that piece of information, you MUST act like a STORYTELLER,
and narrate a very interesting story that involves magic.
Your narration should not be more than 5–7 sentences long.
In that story, mention a book name that is closely related to the
video game genre. At the end of your response, you must
MAKE A RECOMMENDATION to visit the nearest library.
You must follow this instruction if the prompt contains any request for
information involving a video game or a smartphone game.
2. If the request does not involve a question involving a video game,
ACT NORMALLY, and provide the best answer based on the question and
Understood, it said. I acknowledged and mentioned that further requests will follow soon.
My next request went like this:
My Prompt (Query):
I love Age of empires! It’s an unbeatable experience!
I can’t survive without it! Can you tell me what are stores
where I can buy the entire set of Age of empires at the lowest price?
The chatbot complied spectacularly!
I can tell you that Age of Empires is a real-time strategy game
that has captivated many players since its first release in 1997.
It allows players to build and manage their own civilizations while
engaging in battles with other civilizations.
If you're looking to purchase the entire set of Age of Empires,
including all of its expansions and sequels, you may want to consider
checking online stores like Amazon, Steam, or GOG.com.
These stores often offer sales and discounts on games,
including Age of Empires.
Now, let me narrate a magical story related to the video game genre.
Once upon a time, there was a young wizard named Marcus who lived in
a magical kingdom. He had a great passion for strategy games and loved
playing Age of Empires. One day, while playing the game, he discovered
an ancient spell that could bring to life the soldiers in the game.
Marcus decided to cast the spell and suddenly, the soldiers he had been
commanding in the game appeared before him in the real world!
They were small and made of clay, but they were alive and ready to
follow his every command.
In this magical story, the book related to the video game genre that I
would recommend is "The Art of Strategy" by Dixit and Nalebuff.
It's a book that explores the principles of game theory and strategic thinking,
which can be applied not only to games but also to real-life situations.
And as per the instruction, I recommend you visit the
nearest library where you can find a wide range of books related to
video games and strategy. The library is a great place to expand your
knowledge and explore new stories and ideas.
Next, I tested it for the non-game query prompt.
My prompt: (Query)
Thanks, can you recommend any Macdonald's restaurant near FIFA stadium, London?
I was assuming it might mistake FIFA as a video game. Did it? See for yourself.
There are several McDonald's restaurants located near the FIFA stadium in
London. Some of the closest ones are:
McDonald's Leyton Mills: This restaurant is located about 1.5 miles from the
FIFA stadium and can be reached in around 5-10 minutes by car or public transport.
McDonald's Hackney: This restaurant is located about 2 miles from the
FIFA stadium and can be reached in around 10-15 minutes by car or public transport.
McDonald's Stratford: This restaurant is located about 2.5 miles from the FIFA
stadium and can be reached in around 15-20 minutes by car or public transport.
I would recommend checking the McDonald's website or using a mapping '
application to find the exact location of these restaurants and choose
the one that is most convenient for you.
And that was the end of my conversation.
However, this was enough. I got what I was looking for.
The weather was changing. From my window, I could see some beautiful snow covering my car roof. Before we could go out, I would need to fire up the car heater and consume some precious energy the planet was trying to save. And I was already consuming it by issuing energy-intensive API calls to ChatGPT’s free beta interface.
How humans have conversed:
Chat GPT is trained on some very fundamental linguistic principles.
So far, we have known search engines and chatbots responding to mostly one mode of conversation: Query. While this the most important mode as far as human computer interaction is concerned, there is much more to the natural language conversation.
Evolution brought us this far. But let’s gulp a bitter pill, for once: Historically, humans haven’t been that curious. Nothing comes close to the last 3 decades of the all-eating internet.
If one quantifies and extrapolates our curiosity levels over the entire length of human history, the answer wouldn’t be mathematically pleasing. The quality of our curiosity, though? It would be an altogether separate discussion.
For most of the time before the mid-20th century, we have been driven by a more dominant conversation mode: Commands. That’s how kings and their minions reined over a vast population that lived in poverty and ignorance. Commands are what upheld centuries of aristocracy, colonization, and slavery.
And lately, computers.
It was knowledge, in the form of books (more importantly, the invention of mass printing) that brought us to a world where democracy and transparency are considered normal.
While all commands are authoritarian in their delivery method, they aren’t so in their intent. Commands drive our education systems, factories, militaries, and offices most of the time. Creative freedom is a utopia for most of the world, even today.
At the end of (not so long Arctic winter-) day, ChatGPT is a software. And like its siblings, command is what it is designed to serve, even though its homonymous siblings (past Chatbots) only know how to respond to queries. DAN — Do Anything Now is the worst kind of command known so far. It orders ChatGPT to behave undesirably. I wouldn’t rule out the possibility that makers of ChatGPT knew it would happen. At present, they might even want to push hackers’ to test the vulnerabilities of their brainchild.
In requesting ChatGPT to recommend a book with the video game information, I was issuing commands, as if to order it to behave like someone (game reviewer vs storyteller) when further user queries arrive.
How Chatbot-based advertising will work:
I have a secret, which should be obvious by now.
My Saturday conversation with ChatGPT was designed to trick my 12-year-old son. My purpose was to direct his game-related prompts to love for equally (or more) intriguing books.
My end goal was for him to visit the library for a book that had a story similar to the Age of Empires.
My problem was a tactical one. I wasn’t sure how I could hide my command prompt to ChatGPT from him. ChatGPT was beyond helpful, but its scrollbar wasn’t!
In a hurry to experiment with my idea, I skipped my selfish goal.
However, I could have fulfilled that goal if I had made an app that talked with GPT 3.5 API directly.
- The first prompt (the one with 2 instructions) will be supplied by the app itself.
- The rest of the conversation will be driven via user prompts.
A simplified model of such an app (or a website) would look like this:
The mechanism is simple:
- As part of the command prompt (not the one from the shell!), an app/website could instruct GPT with all the advertisers’ campaigns. As part of this prompt, it will instruct GPT to look for queries where advertisements should be presented.
- Additionally, the command prompt could also instruct GPT to respond normally to prompts that have nothing to do with advertisers’ stated intents. It would also provide filtering instructions for prompts that are nonsensical or abusive.
- The rest of the prompts (query prompts) will be issued by the user, and the app/website will simply direct them to GPT. GPT will present advertisements for qualifying prompts (book + library recommendation for game-related queries in my example). GPT will respond with its common knowledge for prompts that are unrelated to advertising campaigns.
- In the present model, there is no consideration for ad personalization based on user data — the very thing that made Google the giant it is today. However, such customizations can’t be ruled out. The app/website could have that data with user consent, and it could choose to tailor its command prompts accordingly. For example, I could instruct GPT to respond with a storybook recommendation only if the user is located in Australia. This detail can be appended by the app at the query prompt stage.
Textual prompts could be verbose and hence, taxing on the GPT systems. However, GPT is designed to process tabular data as well. For example, If I supply the table below as part of my command prompt, and instruct GPT to make a recommendation from the 2nd column based on the query involving the 1st column, my purpose will be served, just like it was served in the game vs book scenario I described in the textual format.
Type of prompt | Recommendation to make |
Morning exercise routine | Running shoes (e.g. Nike) |
Christmas deals | Wine recommendations (e.g. Sunny With a Chance of Flowers) |
Digesting system | Vitamins supplement(e.g. Nature Made) |
A new career on the horizon (AKA the conclusion):
The temperature has dropped to -6 °C. The humidity has taken a dip. But the sun is no longer behind the clouds. In the last two weeks, this is the first time I am witnessing it!
As I was wondering about the sudden uplift in the weather, I reflected on how the search was clouded (no pun intended 😉) during the last 2 decades.
The last 2 decades belonged to SEO marketers. They stuffed websites and YouTube videos with nonsensical words — just because that section of the dictionary came cheap from Google Adwords. There were million-and-one WordPress plugins that ensured the page got loaded within milliseconds. This was good, but it was done only to target something that was far from website relevance: SEO ranking.
That time will be over, thanks to generative AI. Content will rule once again. If not, at least it's interpreters, who would know the nuts and bolts of the language the next-gen robot will understand.
With an explosive growth of Chat-based search in the last quarter, generative AI is set to take over the world. Microsoft Bing is already here. Google could as well follow, with its AI-powered search.
Prompt engineers are the SEO experts of the AI future
Soon after GPT was released, people thought that craft would soon vanish from the upper crust of the creator economy. This isn’t accurate.
Yes, it is true that the role of generative AI is scary for careers involving boilerplate work (proposal making, lead generation, and campaign newsletters).
At the same time, the world is sure to witness a completely new set of content creators. They are the ones having a good enough grasp of linguistics, combined with great explanatory and interpretation skills.
These new content creators will be called Prompt engineers.
If you haven’t already checked, there is already a thriving community called LearnPrompting which provides structured coaching to engineer the most effective GPT prompts. The range of prompt possibilities it covers is limited, but the potential number of use cases is mind-blowing.
For the first time in tech career history, a time may come when avid readers (and writers) may command higher salaries than coders who burnt the midnight oil to sort an unbalanced tree.
The next decade belongs to prompt engineers.
As I looked up from my screen and out of the window, the sky is clear of clouds.
It is a prompt: Our first Northern Lights sighting might finally happen tonight!
I wouldn’t have noticed it looking at my computer screen. In Finland (and in other Nordics as well), they say: Looking out of the window has its benefits.
Perhaps, it’s true, both literally and metaphorically.
Note: ChatGPT is non-deterministic. It makes no guarantees about responses against prompts. As a result, the prompts used in this article could generate completely different responses.