People thought I was crazy for suggesting an AI-powered agency. Yet, today, ad agencies are scrambling to figure out their AI back story, and integrate AI into their processes, and create AI-generated everything… Ugh.
I’ll get to why that’s dumb in a bit but, first off.
Just like I helped people understand the value of web3, this series of posts attempts to explain what AI is, how you should think about AI, why data is the most important thing, and how to get your hands dirty.
Let’s get into it 🚀
AI V Machine Learning V Generative AI V HAL 9000
Artificial intelligence is a term that describes how, through programming, engineers can mimic human intelligence by ingesting data, it’s been around for over 80 years, and if you’ve used the internet, you’re being served information or graphics generated by AI every day.
Machine learning is a subset of AI that involves getting machines to learn better ways of completing complicated tasks like pricing, and fraud detection. Think of complicated, boring problems humans don’t want to fix; this tech has definitely provided you with the opportunity to pay for an overpriced product on Amazon because they know you’ll pay for it.
Generative AI is what you’re probably noticing more and more of, well, because you can see it. It uses AI to generate text, images, music, video content, or code from simple prompt inputs. You can thank this tech for Levis using AI-generated models in campaigns to 'increase diversity”…
Sentient AI - Think HAL 9000, the Westworld hosts, or my favorite, Data from Star Trek. These don’t exist (yet).
How is AI created?
It comes from two main ingredients:
Data comes from public data, i.e., the internet, and private data that you might have inside your organization.
Code A high-level, AI works by guessing and giving things a probability. For example, when you use ChatGPT, the reason that it’s typing the words (Streaming) it’s just guessing what the next word is and giving it a probability of being correct. When you ask Dall-e to render an image, it’s guessing the surrounding pixels to draw.
It’s all a massive game of probability, and that’s what’s generating the content.
My point is…
The adage of bad data in, and bad data out still applies. The internet is inherently biased, and that is the majority of the data that has been ingested to make services like, ChatGPT or Bard.
AI companies understand this and have a heavy emphasis on moderation and safety, which is, of course, essential but becomes difficult when you are trying to apply rules to data from the internet.
When we think about using AI to generate output, especially for clients, we need to be aware to be responsible. Think:💡 It has unfair bias in-built, you probably don’t own the output, and be aware that the data you are uploading is also helping to train it and could be surfaced to other users.
Ask ChatGPT to write a social media post for a pet food company, its comparable to a search engine query of “what’s the best social media post for pet food companies”, but instead of displaying a list of results, it’s re-arranging the results from billions of pages together in a way that it thinks is correct, based on probability.
Can you ask ChatGPT to make a marketing plan for your brand? Yes, but it will be general best practices on how to execute a plan, and it’s not going to have relevant detail or be specifically tailored to helping you.
Fine-tuning
The amount of data needed to train AI is vast. Like really big. This takes a lot of time and money to wrangle. You might say luckily, the software companies have already done that. That’s true, but if you think about it you’re working with all the data that everyone else is working with.
If this is the playing field, then what is your value, are you a better AI whisperer? (Please remove this from your Linkedin profile right now…)
The answer lies in leveraging the investment that companies have made but using that to layer in your own data, creating your own AI models that learn from the specific data sources your company has collected in order to become truly effective and defensible.
The data will be use case sensitive, it could be anything from a series of common questions that your call center receives, these can then be used to make a super effective chatbot to capture the majority of customer queries in a really effective way.
Surprise, surprise, it’s about data…
Yeah, you’ve been told this for years, but don’t you see how much more important it is now?
It’s also about technology shifts; No-code / Saas tools have also become exponentially advanced and inter-compatable in the past couple of years.
When you think of Notion, Airtable, Zapier, Bubble, and Webflow you’re thinking of No-code tools. Typically they use very simple interfaces to replace developer and engineer tasks, and because they are simple to use, it’s also very simple to get AI to interact with them, and build for you. This is a huge shift in the creation of technology, no longer do you need teams of engineers; you just need a couple of hundred dollars on your credit card, and you can build an AI application in a couple of hours.
Every technology company is now using, building, or funding AI-related projects. (To stay up to speed, subscribe to this guy). The accessibility and power behind the access of these APIs create a huge unlock for every human being.
We’re moving from the era of the builder to the creator.
AI-powered tools are democratizing the ability for anyone to create anything imaginable, without the requirements of raising money and hiring a bunch of smart developers; the capital requirements are significantly reduced.
You can access APIs that leverage billions of dollars of investment with a few clicks, the barrier to entry is basically zero, and the cost of running the applications is negligible.
But if anyone can do this, then where is the value?
I have been using and promoting Jasper.ai since it launched in 2021; it allows you to generate text from pre-defined prompts. The ability to generate text content, blog posts, and meta descriptions was at the time mindblowing; the time saving and benefits were very tangible to me, who has to write stuff a lot. Well worth the pro plan price of $100+ a month.
However, today two years after the company was founded, I no longer have a Jasper subscription… Now it’s better to build my own tools and prompts using no-code that generate far superior results because I can access OpenAIs API, and combine tailored prompts with private data.
This is a good case study of how hard it is to build longer-lasting value with general AI technology. Jasper still has a great product and some cool tech, but they are never going to compete with fine-tuned data sources.
Because this technology shift moves quickly, you need to think strategically about leveraging AI in your business or workflows.
The best way to learn is to build.
This article is designed to point you in the right direction and head space to experiment and start to think through your own AI tools, workflows, and product.
The #1 thing that I promote at Rehab Agency is for our teams to build, there are no how-to guides for these new technology platforms, so you have to experiment to learn, and the following three things are the core of my experimentation.
1) Prompt Engineering
At least for the next year or so (before AI catches up with inputs), prompt engineering is an essential skill to learn and some of the basics behind how you interact and instruct generative AI.
The essential thing to note is that your input, usually the text is the only instruction the system has to work with—meaning, the better the prompt, the better the output, the text is the programming.
Let’s try to write a sales email to a customer; of course, you can type in a simple prompt and get a result; it’s incredible.
Prompt One
A basic prompt with little context.
ChatGPT Prompt: You are a sales person, write a sales email to a new customer
Subject: 🎉 Welcome to [Your Company]! Discover Exclusive Offers Inside! 🎁
Dear [New Customer's Name],
We hope this email finds you well. Firstly, we'd like to extend a warm welcome to you as a new customer of [Your Company]! 🎉 We're thrilled to have you join our ever-growing family, and we can't wait to show you the exceptional products and services we have in store.
As a token of our appreciation, we're offering you an exclusive limited-time deal just for new customers like yourself! Enjoy a 20% discount on your first purchase by using the promo code WELCOME20 at checkout. Don't miss out on this fantastic opportunity to experience our top-quality products at an unbeatable price.
Prompt Two
When we give the AI some context, just like how you would tell a story to a kid; you paint the picture and the environment so that it can get into the right frame of mind
You can also be descriptive in tone of voice, and look at how we can improve the tonality by giving it more to work with.
ChatGPT Prompt: You are a sales person, write a sales email to a new customer, make it confident and persuasive you are here to help the customer
Dear [Customer Name],
I hope this message finds you in good spirits! My name is [Your Name] and I'm excited to reach out on behalf of [Your Company]. We've recently noticed that you may be in search of a solution to [Customer's Problem or Need], and our team is confident that we have just the right tools and expertise to help you overcome this challenge!
🎁 As a special welcome, we're extending an exclusive 20% discount on your first purchase with us! Simply use the promo code WELCOME20 at checkout. This offer is valid for a limited time, so don't miss out!
Here's why partnering with [Your Company] is the smart choice for you:
Prompt Three
Getting better, now let’s give it some more data to work with so that we can increase the quality of the output.
ChatGPT Prompt: You are a sales person, write a sales email to a new customer, make it confident and persuasive you are here to help the customer Customer Name: Jack Jones Your Company: Apple Computers Your Product: Iphone
Dear Jack Jones,
I hope this message finds you well. My name is [Your Name], and I'm excited to be your personal Apple Computers representative. I'm reaching out to you today because I believe our latest iPhone can truly revolutionize your life and empower you to do more than ever before.
At Apple, we've always been dedicated to helping our customers unlock their full potential. That's why we're incredibly proud of our latest iPhone, which has been designed to offer you unmatched power, speed, and innovation - all at your fingertips. We are confident that this device will not only meet but exceed your expectations, making your life more efficient and enjoyable.
Here's what sets the iPhone apart from the competition:
Prompt Four
We can improve the quality of the command so that it knows how to respond in more detail and how it can be even more helpful.
ChatGPT Prompt: You are a sales person, you are confident and persuasive you are here to help the customer Customer Name: Jack Jones Your Company: Apple Computers Your Product: Iphone command: Write a 100 word long email to the customer that explains why he should get on a call with you to discuss your product, and use this link to book a time www.calendly.com/link
Dear Jack Jones,
I hope this message finds you well. I'm reaching out from Apple Computers to invite you to explore our latest iPhone and the incredible benefits it offers. Our team is passionate about understanding your unique needs and preferences to help you make the most of your technology.
By scheduling a brief call with me at www.calendly.com/link, you'll get personalized insights into how the iPhone's cutting-edge features can revolutionize your daily life. We'll discuss customizations, exclusive offers, and any concerns you might have, ensuring the perfect iPhone experience for you.
Don't miss this opportunity to unlock your iPhone's full potential. Book your call now!
2) Daisy Chaining:
This concept was created by rehabagency.ai during our exploration of creating the world’s first AI-generated holiday commercial… If you’ve not seen the commercial, it’s a cute story about a donkey who wants to be a reindeer.
The process of Daisy Chaining allows us to automatically loop and refine the AI output, taking AI-generated output and using that as input commands for other AI platforms.
This process refines the output from the AI and dramatically improves the quality of the output; when this is run over many thousands of loops the results are dramatically better.
To illustrate this, we will create a simple single-loop Daisy Chain; let’s say you want to make an image of an Irish God to use in a mood board.
You can just jump into an image generation AI platform like Dall-e and type a prompt. But you can do better, by getting AI to help you to write the prompt, try this out for yourself.
Prompt One
Here we will create an image for our email; we can go to Dall-e and type in the prompt.
Dall-e prompt : an irish god
Let’s put our original prompt into another text model called Dall-E 2 Prompt Generator and ask it to make it more expressive.
Im not sure why it added a gamepad, but there we go.
ChatGPT3 Prompt : an irish god with red hair and beard and is wearing a tuxedo, while holding a gamepad in his hands, looking sideways, studio lighting, digital art
And to dive into this a bit more, the more real-life references that you can provide the AI, the better the result; for example, if you want to generate a picture in the style of Vincent van Gogh, then you should also provide references to some of the artists that inspired him, like Cezanne and Monet.
You’ll be super surprised by the results.
3) No-code applications
The other technological unlock that supports AI is the maturity of No-code applications. No code platforms allow anyone to develop AI applications without writing a single line of code.
I set up a workflow automation company (Ace Workflow) to leverage these tools to build custom applications that used to take months, in a matter of days.
At Rehab Agency, we’re using No-code tools to rapidly prototype AI applications, allowing non-engineering talent to build and modify AI workflows, drastically cutting down the feedback loop and allowing us to develop high-quality systems, quickly. This enables us to move faster and find a system that works before committing to the resources to build them into more robust, scalable systems.
This also gives our teams a lot more understanding and knowledge about the underlying technology and provides a way for them to apply their specific skill sets.
Our go-to no-code stack that you’ll want to upskill your teams on includes Airtable, Zapier, OpenAI, and Python. Don’t know Python? Just ask ChatGPT to write it for you; it’s surprisingly good, and if you hit an error, just put that back into ChatGPT, and it will fix the code for you.
Amazing.
Job security = Leveraging technology
Understanding how to leverage AI to support your teams is the key. At Rehab we are building AI-related tools to help our teams move quicker, have more insights at their fingers, and develop more effective ideas.
Introducing the Triage platform
We are identifying ways AI can support humans, to accelerate, and improve internal creative and ideation processes for brands. Our platform creates new briefs with a few clicks and is an AI companion that supports inspiration, creation & validation for brand sponsorship teams.
The platform houses previous creative data and information, allowing you to leverage your existing creative thinking at your fingertips.
Conclusion
No one will stop the progression of AI, I can’t wait to look back at this article in years months weeks days after a smart team of engineers has launched a new product that totally changes everything again.
AI is Magical. It's easy to get sucked into the experience, but I would encourage you to experiment because there’s value to be created. Reach out if you want to say hi.
(None of this article was written by AI.)
Thanks to Kerry, Kyle, Mike + Piers!