I recently set out to learn everything I could about AI with the goal of figuring out how I can apply it to existing businesses. Here’s what I learned.
This is from a developer’s perspective focused on how to integrate AI into existing products to make them better.
As I am a learn by doing kind of person, I built little AI apps along the way that showcase various uses of AI. You’ll be able to try them out and maybe you’ll find them slightly useful. I built them as chat bots on Telegram as I’m a fan and have built many Telegram bots over the years. The chat style interface is a great fit for AI applications and is likely how you’ve already tried AI in ChatGPT or Gemini for example.
So what are the main use cases for AI today? Let’s go through them.
I tried image generation first as it is the most fun and visually appealing part of AI. If you haven’t tried it before, it’s quite magical the first time you do. You think of an image you want, you type a description of that image in natural language and boom, in seconds you have yourself a new image. And sometimes these images are incredible.
The clearer you can describe your image, the closer that image corresponds to your idea. Most likely it won’t be exactly what you had in mind, but you can usually get it good enough, and it might even create something better than you anticipated.
Here are some examples of some really amazing images:
Ok great, but you can generate images using any AI tool right? Right, but if you’re on Telegram, AI Monster is for you. The fun part about it is the social aspect, anyone in a group can quickly generate one-of-a-kind images and everyone can see it happen in real-time. Challenge your friends to see who can create the best image.
There are a ton of ways to use this and you’ve probably seen it used it in some way or another.
You can ask AI things like “write me an essay about X” and boom, in seconds you have a fairly well written document that might have taken days otherwise.
But this stuff is table stakes now, so what are some good use cases other than cheating on your schoolwork?
A common use case you’ll see everywhere now is AI customer support chat systems. You click on the chat icon on a website and start chatting. Prior to AI, it was either a human you were interacting with or some pre-defined tree of choices. Now you can talk with natural language and get natural language responses that feel like you’re talking to a human.
This is powerful when trained on prior support data and knowledge base data (see training section below).
A surprisingly useful and good use case is summarizing. Take a bunch of content and sum it up in a sentence or two. Think of it like a modern day Coles Notes or CliffsNotes; if you’re too lazy to read the whole thing, just read the notes!
You may think this doesn’t warrant a whole category, but I beg to differ. We’re going to start seeing this everywhere and it’s already in a lot of places such as Amazon review summaries and everywhere in Google Workspace.
An actual Amazon review summary for socks.
Very useful and not the end of the world if it gets it wrong.
YouTube is even starting to do with this videos. The screenshot below is a 20 minute video condensed into 3 sentences. Good for me, bad for creator.
Press enter or click to view image in full size
Search is another great use case. Google is doing it already at the top of your search results. Amazon is doing it for shopping with their newly launched Rufus AI.
The basic idea for AI search is:
Search for something or ask a question
Get an answer in natural language
That may sound like normal search, but it’s not. Current search is:
Search for something or ask a question
Get a list of potential results that may contain the answer
Click through those results until you find the right answer
Subtle, but HUGE difference.
I do believe AI will replace search as we know it. It’s just simply a much faster way to get an answer. And it can still return a list of results too if you need to dig in further. So just generally an all around better search experience.
The customer support stuff above is basically the same thing as search. Ask a question, search knowledge base, give me an answer that sounds like a human is talking to me.
I think we’ll see custom trained AI models replacing things like Elastic Search which is the defacto standard for search technology today.
I think this use case might actually take off, mostly because it’s fun and can be great entertainment, but also because it may end up being quite useful.
Join Medium for free to get updates from this writer.
Subscribe
Subscribe
There are also options like Girlfriend, Boyfriend, Therapist and Life Coach. I had my wife try it out:
I don’t think she’s using it right. :)
This is one of the most interesting areas of generative AI, and it is likely the most promising area for business use. The basic idea is that you feed the model your own data so that it can learn from it then generate results based on that data.
I’ll use training/fine-tuning interchangeably as they are similar concepts. Fine-tuning is basically training on top of an existing model. Training (or pre-training) is training a model from scratch. You’ll likely use fine-tuning for most cases so you can take advantage of the hard work others have spent on training huge models and just layer on top of it.
For text, you can train on your own internal/private data. A common use case for this is to use customer support/documentation/knowledge base data. When when a customer has a question, the AI can attempt to respond based on your own data in a way that feels like a human responding.
For images, you can train on various aspects of images. I tried a couple of things that worked fairly well and came up with some interesting results.
First I tried training a model on my face, then I was able to prompt my custom model and put myself in various scenarios like this:
lol
What did I learn? I learned that I can make myself look way cooler than I am. :)
Another thing I tried is more business related. I wanted to see if I could generate images that could be used for marketing and branding. So I took a logo that contains two colors, blue and pink:
Then I was able to generate themed images that match the brand colors:
Press enter or click to view image in full size
Now this technology could definitely be used for marketing materials. In seconds, I can generate images that match up with a companies branding.
If you’re in the creative industry whether that’s marketing, gaming or entertainment, then I think you have to start using AI. It would be very unwise to ignore it as it can drop the cost of many tasks by at least an order of magnitude and everyone wants cheaper things. These areas are in for a fundamental shift. My advice would be to embrace it though, rather than be scared of it as you can’t stop it. The genie is out of the bottle. Learn it, use it, figure out how it can make you better and faster at what you do.
I am in the exact same boat as a software engineer, AI can produce pretty decent chunks of code. I hear a lot of chatter about AI replacing developers, but it won’t because it doesn’t really think and it doesn’t know what you want to build. That said, I have fully embraced it to make me a much faster developer. I use these AI coding tools all the time as part of my development process and it’s actually really great.
Should you compete in general AI. Absolutely not! You’ll spend billions and still probably be left in the dust by the biggest companies in the world doing the same thing. This will become commoditized and arguably already has. It’s a race to the bottom and they’re all training on the same data so they’ll all end up roughly the same anyways. What’s the point?
Do you have a bunch of proprietary data that you think could provide valuable insights with natural language questions and answers? If so, then t his would be a good place to start experimenting. Train your own model on your data.
Keep in mind, AI is not good at math or providing exact answers so the uses for this are limited. If your business has a lot of numbers and graphs you want to expose that need to be exact, don’t use AI (yet).
Should you buy a bunch of GPUs and train models yourself. No. Capitalize on the companies that already bought a bunch of GPUs and let them feel the pain while you take advantage of the hard earned dollars they’ve already burned.
What if I don’t want my private data on their GPUs? If you are already using a cloud provider, does it matter? Use their GPUs. Every cloud provider has AI services now.
Should you add AI to your marketing materials? Yes! Your competition already did it and you don’t want to be left behind. If someone asks how you’re using AI, say it’s a trade secret. If your company name is Dog Food Depot, change it to AI Dog Food Depot. And if your tagline is “Organic, free range, no filler dog food”, change it to “Organic, free range, AI enhanced dog food”. This will increase your valuation by at least 5x.
In all seriousness, I would proceed with caution if you’re planning to use AI. It’s very expensive and it’s far from perfect. Every company is going to pitch you that it will change your entire business operations, that you can fire your entire staff, etc., but keep in mind the worst problem with AI: you don’t know what you’re going to get.
AI is like a box of chocolates, you never know what you’re going to get.
You can take the same input and every single time it will generate a new output. Whether that’s a new image or new text.
This means your AI will often be wrong and you need to ask yourself how much wrong you are willing to accept.
Imagine your customer support AI bot delivered the correct answer 95% of the time. Sounds great right? It’s not. You’re going to really piss off 5% of your customers.
There are a lot of cases where it doesn’t really matter if it gets things wrong and these are areas you should likely focus on in the short term. Like summarizing sock reviews on Amazon.
Longer term, this technology will get better and that rate of wrong (ROW) will surely decrease. But for now, if you’re going to use it, you need to accept this problem.
I should probably mention the cost here, because it’s not cheap. If you do it all yourself, you’re going to need expensive beefy machines with GPUs. If you use a cloud provider, same thing. If you use an AI as a Service provider, it’ll be more expensive on a per generation basis, but you don’t have to deal with infrastructure or maximizing the usage of that infrastructure.
Each generated image is in the pennies range. That’s not 1000 images for pennies, that’s one single image. Generating text is cheaper, but still insanely expensive compared to what you are probably used to when it comes to infrastructure costs.
If you don’t see a clear path to value by adding AI to your product, I would probably hold off for a while until the hype passes, the real use cases get squeezed out and the price comes way down.
But since I don’t practice what I preach, I’m excited to announce:
If you don’t have AI in your business, you’re dead in water. Want to raise money without AI in your name? Good luck.
Press enter or click to view image in full size
Business, governance, and adoption-focused material. Real-world implementations, case studies, and industry impact.
