Cute, isn’t he? Or is it a her? Do robots have the same rights as humans? Will AI come for our jobs? Will they come from the future to find Sarah Connor? These are all burning questions, but not the topic we’re discussing today.
Writing content can suck sometimes, especially when you’re writing within certain boundaries or where you have to really stretch your imagination to meet a character limit. I hate writing meta descriptions for pages for example. Like, really hate it.
Advances in AI and neural networks have however started to shift the way we approach writing copy for web and other applications. GPT-3 for example has caused widespread fear of an AI revolution with their ability to write content in natural language. I am of course just being cheeky with the pronouns. GPT-3 is not a thinking machine but rather a neural network that uses machine learning techniques to learn natural language and produce content. GPT-3 is just one of the many APIs or neural networks that can be utilised for content generation. Today we’ll be looking at one such AI service: InferKit.
Using InferKit, one of many services
InferKit uses a neural network to learn and write content, and for the most part, gets it relatively right. I couldn’t find any proof that InferKit uses OpenAI’s GPT-3 but it’s highly possible given their comment on Elon’s musical talent when using their Talk to Transformer demo (I’m waiting on that Grimes and Elon collaboration now, thanks guys).
InferKit is simple to use and has free options available for us average Joes – I am yet to use up my monthly character allocation. Creating an account is quick and will have you playing with AI in a matter of minutes. Their service takes some starting text, whether it’s a full sentence or a few words, and spits out additional content based on what it’s learned from the Internet. I’ve found that providing it with more detail does tend to yield better and more relevant results.
Some basic examples
For a few months now I’ve been running an SEO experiment on Snug Cooking, a simple website where I share recipes from my family. Any SEO specialist will know that content is key to growth and generating content for recipes is not easy (you can only talk about how awesome a carb is so many times before it gets stale). This is where InferKit has really come to the rescue.
In this first example, I generated some additional content for a potato curry recipe. The premise of the recipe is that potato curry is affordable and simple. I fed InferKit this starting sentence: Comforting potato curry that’s a cost effective meal. What InferKit generated was perfect; it spoke to the fact that potatoes actually have fewer calories than rice and even cheekily suggested that potato curry is more nutritious than curry with rice. I had to make small tweaks to the final version, and fact check the measurements, but this is what it came to:
Comforting potato curry that’s a cost effective meal. Potatoes have a low caloric content (around 2 kcal per 1 gram) and a low glycemic load, which makes this dish more satisfying and nutritious than if it had been made with rice.
This recipe intro was one example where InferKit got it right from its first try, but it doesn’t always hit the mark. For this focaccia recipe, I fed it a bit more context than I needed to and the results were not necessarily useful. I also generated text a few times, each attempt having varying levels of accuracy in the results.
What was useful about this exercise though is that it, at times, generated content that could be used for other recipes. In this example, I could quite easily use part of this content for my pizza dough recipe:
The only change I’ve made to this recipe is to fry instead of bake it, which makes it even more simple. So many rich, wonderful flavors come from this delicious and versatile pizza dough!
What about longer pieces of text?
InferKit worked relatively well for smaller snippets of content, even though it required some tweaking and patience at times. How does it deal with longer pieces of content though? In this example I gave it the intro of my article on how a phishing scam works. Here’s the introductory snippet I fed into the algorithm:
Phishing scams get more sophisticated by the day and as a result it is so important to understand how these scammers operate.
What it spat out really surprised me, it was a great primer on the basics of phishing scams and how they work and I actually might as well have written the entire thing with InferKit by the looks of it:
I wanted to write this article to educate the wider community in the security industry and consumers about how phishing works and how to recognise a phishing email. How do Phishing scams work? Phishing scams work by sending phishing emails pretending to be from a trusted company or person (for example, your bank) to your email address.
Closing thoughts
I wanted to close off by telling you my secret. This entire article was not generated by an AI, sorry to disappoint you. There is also no secret, but I hope you enjoyed the suspense.
It’s safe to say that we are quite a way off from having terminators on our doorstep and services like InferKit are really only the first step in automation through AI. With that said, we still need to use it responsibly, as search engines place more and more emphasis on original content and its role in providing our readers with a great experience. I end off with a very familiar quote: With great power comes great responsibility. It’s undeniably corny but holds even more true as we usher in the era of the machine.
