GPT-3 in Action: Writing Apps Based on OpenAI’s Language Model
GPT-3 powered writing apps can produce incredible, human-like copy and content. We tested how they work and why average users prefer them over OpenAI’s GPT-3.
GPT-3 is OpenAI’s autoregressive language model. It can produce human-like text, understand natural language, and even interpret programming languages in simple, natural-language terms.
OpenAI released GPT-3 API in June 2020. Since then, many companies have integrated GPT-3 with other technologies. This enabled average users to perform various complex tasks using AI, such as writing content or code.
GPT-3 powered apps make the technology more accessible to mainstream users. OpenAI’s GPT-3 is not very user-friendly or intuitive, which makes it difficult to use for non-technical users. GPT-3 powered apps, on the other hand, are far more intuitive and easy to use, which is why they’re more popular with the general public.
Copy.ai is a GPT-3 powered app specializing in writing marketing copy. It can write blog posts, website copy, product descriptions, emails, social media posts, and more in just a few clicks.
Lex is an AI writing tool and word processor. It uses GPT-3 to help users write content and do research in half the time it usually takes.
The biggest advantage of GPT-3 powered apps is that they’re specialized. Instead of targeting everyone, GPT-3 based apps help specific users solve problems tied to one niche or industry. We expect the apps to become even more specialized in the upcoming years.
GPT-3 may be used for questionable purposes in the future, such as chatting with deceased loved ones. This will surely spark further controversy and debate about the ethical use of this powerful technology.
This post is sponsored by Multimodal, a NYC-based development shop that focuses on building custom natural language processing solutions for product teams using large language models (LLMs).
With Multimodal, you will reduce your time-to-market for introducing NLP in your product. Projects take as little as 3 months from start to finish and cost less than 50% of newly formed NLP teams, without any of the hassle. Contact them to learn more.
Ankur’s Newsletter is a reader-supported publication. To receive new posts and support our work, consider becoming a free or paid subscriber.
It’s been two years since OpenAI first released API access to GPT-3. Since then, companies have rolled out one GPT-3 powered app after another—and the user feedback has been extremely positive. As an example, consider one of many glowing reviews of Copy.ai, a GPT-3 powered writing assistant:
CopyAI is a marketing writer's dream. It's like having a ghostwriter, but without the expense and the hassle of having to hire and manage someone else to write your content.
The rise of Copy.ai and similar apps allowed average users to finally start benefiting from GPT-3. They were suddenly using it to perform various complex tasks, such as writing content or code, in half the time and with half the energy.
In this post, we’ll try to discover how that happened and how GPT-3 based apps differ from OpenAI’s GPT-3. Is there something that makes apps more suitable for average users than OpenAI’s out-of-the-box GPT-3?
To answer that, we’ll analyze two GPT-3 powered writing assistants, Copy.ai and Lex, and compare them to OpenAI’s tech. But first, let’s quickly explain what GPT technology is.
What Is GPT Technology?
GPT (generative pre-training) is a deep learning method developed by OpenAI. It involves training language models on large amounts of data in order to improve their ability to predict the next most probable word in a sentence—or the next most probable sentence, paragraph, and so on.
GPT-3 is the third and last version of OpenAI’s autoregressive language model. It’s also the largest natural language processing (NLP) transformer to date, as it’s been trained with over 175 billion parameters. Thanks to that, GPT-3 can now perform complex linguistic tasks like translation and content generation stunningly well.
In short, GPT-3 provides natural language text completions based on user text prompts. For example, entering a question can work as a text prompt for an answer to that question:
Now that we have an understanding of what GPT-3 is, let’s see how companies integrated it with their tech to develop new use cases.
GPT-3 Commercial Use Cases
OpenAI released API access to GPT-3 in June 2020. Just nine months later, the company announced that over 300 apps use GPT-3 for various purposes: content generation, code interpretation, code writing, SQL request generation, and much more.
Since content generation is perhaps the most talked-about use case of GPT-3, we chose to test two AI content writers, Copy.ai and Lex, in this issue. We’ll analyze them one by one and then try to conclude how they differ from OpenAI’s GPT-3. Let’s dive in.
Copy.ai is an AI tool that specializes in writing marketing copy. It can write blog posts, website copy, product descriptions, emails, social media posts, and much more in just a few clicks.
But it also has a few other interesting functions. For example, it can help business owners generate new product ideas, product names, value propositions, and other business-related ideas. It can also help users write content for personal use, such as cover letters, birthday cards, and even “shower thoughts.”
The following sections will explore how these core features work in more depth.
Copy.ai is primarily a writing tool. It can write numerous types of marketing content and copy, some of which we’ve already mentioned:
Website copy — Separate tools for different parts of website copy (e.g., calls to action, meta descriptions, microcopy, subheaders, etc.)
Blog content — Blog post generation tools (e.g., blog outlines, intros, outros, etc.) and Content marketing tools (blog ideas generator)
Ad copy — Tools for generating ad copy for different marketing channels (e.g., LinkedIn, Facebook, etc.)
Emails — Tools for generating emails for different purposes (e.g., follow-up emails, thank you emails, welcome emails, etc.)
Sales copy — Tools for generating sales copy based on various copywriting formulas (e.g., PAS, AIDA, 4P’s, etc.)
Social media copy — Tools for generating ideas, hashtags, and emojis and Tools for text generation (captions, descriptions, and titles)
Product descriptions — A product description generator
The AI creates content or new ideas based on user entries (i.e., text prompts). For example, here are our entries for this blog post:
The AI created several outlines for our blog post based on this entry. Here’s one of them:
The AI-generated outline looks quite similar to the actual outline we used for this post. However, to be fair, we gave the AI very detailed instructions, which helped it get the outline right.
Let’s see what happens when we give it a simpler prompt. Will it still create a quality outline? You can judge for yourself:
This is just one example of what Copy.ai can produce from simple text prompts. It actually creates several outlines and content options based on one prompt, so you can choose the one you like best. Of course, if you don’t like any of your results, you can always generate more.
Copy.ai has many more writing tools, but we can’t explore them all in this post. So, here are just a couple more examples of what Copy.ai can generate from simple prompts:
Personal and Business Non-writing Tools
Copy.ai also has several useful non-writing tools. Some of them are dedicated to businesses and startups. These include the Name Generator, Motto Generator, Value Proposition generator, Audience Refiner, and a few other tools.
Their purpose is to help businesses grow and brand themselves properly.
Other tools are more targeted toward personal use and are very creative. For example, the Birthday Card tool helps users come up with original birthday wishes, while the Love Letter tool helps them woo their ideal partner.
Lex is a writing tool conceived as an alternative to Google Docs and similar non-specialized word processors that are often inadequate for professional writers.
However, we won’t discuss the differences between Lex and Google Docs here. Instead, we’ll focus on how Lex uses GPT-3 to help writers in the content creation process.
Answering Research Questions
Writers often need to research different subjects and terms while writing. They usually do this in a separate tab, mainly by googling their questions. Then they typically go back and forth between their document and research tab, which can lead to unnecessary friction and confusion.
With Lex, writers can find answers to their questions without leaving the platform or doing too much research. Instead, they can simply enter their questions into the chatbox and wait for Lex AI to answer.
An added benefit is that Lex AI can understand and remember context, which means it can usually give accurate answers even to vague questions. For example, notice that Lex can accurately answer questions like “what kind of AI models” and “what are those” by considering what was said before:
Lex also acts as an AI copywriter, but unlike Copy.ai, it doesn’t have specialized writing tools. In other words, it doesn’t know the difference between blog posts and product descriptions or any other type of copy or content.
Therefore, it also can’t produce specialized content or copy. For example, it can’t create sales copy based on the AIDA formula, which is Copy.ai’s forte.
However, that doesn’t mean that Lex is inferior to Copy.ai. It simply works in a different way. Actually, if we’re being honest, it seems to produce even better content.
Unlike Copy.ai, Lex doesn’t require special prompts to create new content or copy. It simply treats everything a user writes before giving it a command as a prompt.
That means you could potentially write thousands of words and get Lex to consider them all when generating new content. Of course, the words immediately preceding your command will have the most influence over what Lex generates—but the entire document will have some impact, too:
Other GPT-3 Powered Apps
GPT-3 is the underlying technology behind many AI writing assistants. We’ve mentioned Copy.ai and Lex, but there are also Jasper, Frase.io, Zyro, and many more.
However, GPT-3 has more to offer than just pure content generation. It can perform many other tasks, as demonstrated by various GPT-3 powered apps in different niches.
Here are just a few examples of apps that use GPT-3 for purposes other than content generation:
Algolia — Algolia uses GPT-3 to enhance semantic search. Its search technology can now give users better answers to their questions and point them directly to a place on a page where they can find relevant explanations.
Replier.ai — Replier uses GPT-3 to create branded replies to customer reviews. It analyzes the style of previous replies and then mimics it in the new replies it generates.
Quickchat — Quickchat provides businesses with GPT-3 powered AI assistants (read: chatbots). The chatbots can give customers precise and accurate answers that sound human without too much training.
Replit Code Oracle — This GPT-3 powered app acts as a coding ghostwriter, tutor, and consultant. It can write and “translate” codes to natural language and even suggest how an existing code can be improved.
Designer — According to its maker, Jordan Signer, Designer is a text–to-design Figma plugin that allows users to create Figma designs from natural language text prompts. It seems like the plugin has not yet been released to the public.
OpenAI’s GPT-3 vs. GPT-3 Powered Apps
To say that GPT-3 powered apps are popular with the public would be an understatement. Copy.ai claims to have had as many as 3,000,000 users to date, while Lex currently has over 26,000 people on the waitlist. Those are huge numbers for “scrappy startups.”
Interestingly, OpenAI’s GPT-3 didn’t receive the same attention. In fact, average users have probably never heard of OpenAI or GPT-3, let alone used it. On the flip side, most are now well aware of AI writing assistants and may even know apps like Copy.ai and Jasper by name.
That begs the question, why? Why are GPT-3 powered apps so much more popular than OpenAI’s out-of-the-box GPT-3?
The reason is not the quality of output. We tested OpenAI’s GPT-3 against several specialized writing apps and concluded that it can produce the same quality of content.
We believe that some users are intimidated by OpenAI’s GPT-3 and perhaps even the idea of using advanced tech to perform tasks. Writing content with Lex may simply seem less daunting than using something called GPT-3.
Also, GPT-3 powered apps have better UI and UX. They’re more user-friendly, more intuitive, and simpler to use. This has a lot to do with the fact that they’re specialized, i.e., meant to help a specific type of user complete specific types of tasks instead of trying to help everyone do everything.
For example, AI writing apps are primarily meant to help writers and agencies create content or copy. Unlike OpenAI’s GPT-3, they’re not supposed to also help users write code or reply to customer reviews. Again, there are specialized apps for that.
What’s the Future of GPT-3 Powered Apps?
We expect that GPT-3 powered apps will become even more specialized in the future. Niche apps have proved to better solve their target audience’s problems and pain points than apps that cater to everyone. That’s why users enjoy using niche apps more and why niche apps are more intuitive.
So, AI writing apps may target customers in particular niches or industries. For example, one copywriting app may target eCommerce stores that need a lot of ad copy, while another may target B2B businesses that need lots of long-form content.
We can also expect to see more questionable use cases of GPT-3 in the following years. Some, like YOV (You, Only Virtual), have already sparked controversy. YOV is a GPT-3 powered app that lets users chat with deceased loved ones—or, rather, their AI-generated personas. We’ll probably see more similar apps in the future, and we’ll have to discuss what’s ethical and what’s not.
Of course, these trends won’t just impact writing apps. They’ll impact all GPT-3 powered apps, including those that write and interpret code. But we’ll cover that in our next issue.
In the meantime, keep an eye out for new apps based on DALL-E 2. We’ll surely see some exciting use cases of this tech soon, as OpenAI just released the DALL-E API.
Subscribe to get full access to the newsletter and website. Never miss an update on major trends in AI and startups.
Here is a bit more about my experience in this space and the two books I’ve written on unsupervised learning and natural language processing.
You can also follow me on Twitter.