How AI Writing Tools Work
A look beyond the hype
If you’ve been in the SEO community for a while then you already know about the AI writing tool Jasper and its countless competitors.
Most people have no clue how these tools work. In my opinion, it’s always important to get an in-depth understanding of any tool you use so you can better understand how it’ll affect your business. There’s literally no benefit to not understanding something.
In today’s article I give an overview of the technology that underpins modern AI-generated content. I explain how to use it to help your SEO and also go over some of the risks you NEED to be aware of when using this relatively new tool. Let’s get started.
The technology
Generative Pre-trained Transfer 3 (GPT-3) is the technology that Jasper and other AI writing tools are based on. If you aren’t aware, GPT-3 is a language model that generates different types of outputs (text, code, etc.). It uses statistical probability to generate words in a sequence.
GPT-3 was invented by OpenAI in June 2020. It was opened up to the public via API access in November 2021.
When GPT-3 was trained, it used over 45TB of text data using over 175 billion different parameters. It’s estimated that the training data encompassed 10% of available content online.
The dataset for GPT-3’s training consisted of the following:
Common crawl - Webpages, meta data, and text extracts
WebText2 - Pages from outbound Reddit links
Books1/Books2 - Online book collections
Wikipedia - Self-explanatory
It’s estimated that GPT-3 was trained on 3,865,470,525 text pages.
What is GPT-3 capable of?
Creating unique, long-form text
Generating outputs based on a single command
Relatively high-quality output
Understanding grammar, syntax, and semantics
What are the limits of GPT-3?
Accuracy - GPT-3 is not the best at finding factual information
Changing information - Since the algorithm was primarily trained in 2019, it doesn’t have the most up-to-date information about current events.
In-depth information - The algorithm is good at surface level info, but if you’re writing complex technical content then it won’t help much.
What type of content works well with GPT-3?
Common-knowledge topics with static, unchanging information work best with the GPT-3 algorithm.