Photo by Oladimeji Ajegbile on Pexels
In today's world, Artificial Intelligence can generate text that is indistinguishable from what an actual human would write. It’s possible to generate almost any kind of text from blog posts, to dialogues, copy, and even poetry. I will show some examples below.
But is it possible to detect if what you are reading is written by an actual human being or a machine?
Let’s find out.
Large Language Models
The technology that makes all of this possible is Large Language Models or LLMs. These are machine-learning algorithms that are trained on very large data sets of text.
You could think of them as machines that have had access to a large part of the text on the internet and can recognize patterns, structures, and relationships between building blocks of these texts.
Using this data the machine-learning networks can generate new text while following the accepted patterns, structure, and relationships of the words that make up the text.
The most popular LLMs are currently OpenAIs GPT-3 and Google’s LaMDA. But there are plenty more.
Most of the text generation tools out there like Jasper AI or Copy AI are based on GPT-3 or similar text models. They are simply fine-tuning the outputs these models generate.
What’s possible
As I already laid out in the introduction these tools can generate pretty much any kind of text, and even code.
If you would like to learn more about AI-generated code, check out an interview I had with the founder of AskEdith.AI text-to-code tool.
Demo
Now let’s look at some quick examples of text generation. I will be using OpenAI’s Playground which utilizes the GPT-3 model for this demo.
Let’s start with a blog article outline.
This is the prompt I am going to input into Playground:
Generate an outline for a blog article “best online side-hustles to start in 2023”
And here’s what it spits out:
screenshot from https://beta.openai.com/playground
The highlighted text is generated by AI. Let’s make it write the article now.
screenshot from https://beta.openai.com/playground
Bear in mind that I am only using the default settings. And am not being very specific with my prompts. From only 2 short sentences I now have a pretty solid basis for a generic article.
Would I write something like this myself? No!
Is stuff like this out there on the internet? Definitely!
How to detect AI-generated text
There are some instances when we really don’t care if the text is written by AI or a human being. For example, if we are looking for a simple and short answer to a basic question it shouldn’t really matter.
But when we are trying to get more valuable information experienced human expertise is what we are looking for.
While there is no fool-proof way to tell if the text is AI-generated for more fine-tuned generations. There are some giveaways and tools you could use to check if the content is generated.
1. Giveaways
- The text usually lacks a specific tone or character.
- Most of the time the text is written in 3rd person
- Feels plain
Again, these are not dead giveaways. As more fine-tuned generations may differ.
2. Software
You can use this machine learning tool to check if the text is AI-generated or not simply by copy/pasting the text into the tool.
GPT-2 Output Detector
This is an online demo of the GPT-2 output detector model, based on the🤗/Transformers implementation of RoBERTa. Enter…
huggingface.co
It’s not 100% accurate but as far as my personal experience goes, it does a pretty good job.
Implications
LLMs are wonderful marvels of technology. But as with all great inventions, there are some legitimate concerns. With AI-generated text being almost indistinguishable from human-written text, there is a danger of the internet becoming flooded with AI-generated content.
Which would mean ever-increasing amounts of information at an alarmingly fast rate. There is already way more text on the internet than we could possibly dream to consume. And there will inevitably be more. But we have to be careful to protect important information from getting lost in the flood of generic, machine-generated content.
I am not against the use of generative AI tools, but I strongly believe that we need to have a way to distinguish what’s written by humans, and what's generated by machines.
Stay curious and stay
HungryMinded
!
If you are interested in reading more articles on AI-related subjects check out the list below:
HungryMinded
AI and Creativity
View list
17 stories
If you enjoyed reading my work Sign up Here to get all of my future posts directly to your inbox.
Join Medium for just $5 and read an unlimited amount of informative stories from me and thousands of other great writers. You can do that by clicking the link below.
