Key Takeaways

ChatGPT is a conversational language model trained using Reinforcement Learning from Human Feedback (RLHF). Unlike its sibling model, InstructGPT, it can participate in dialogues with users.
ChatGPT products will be more intuitive than GPT-3 products and have fewer limitations. They’ll allow users to interact with apps and devices in a conversational way and make using them more convenient.
We can potentially apply ChatGPT to all industries, from programming to teaching.
ChatGPT is still problematic. Users showed it can produce unsafe content, give incorrect and nonsensical answers, and bypass rules set by OpenAI.

No matter which social network you’re on, news about ChatGPT has probably taken over your feed. Over a million users have already tried OpenAI’s newest language model, and we’ve seen dozens of incredible use cases.

We’ll refer to these use cases and existing GPT-3 powered products to predict future ChatGPT products. Essentially, we’ll try to answer an important question on many people’s minds: where is this all heading?

To kick off, we’ll start with a quick refresher on how ChatGPT differs from similar models. If you don’t need it, feel free to jump straight to the product section and try to imagine the future with us.

A Short Intro To ChatGPT

ChatGPT is a sibling model to InstructGPT. According to ChatGPT itself, it’s a more specialized version of good “old” GPT-3, designed specifically for chatbots and conversational apps.

ChatGPT vs GPT-3, as explained by ChatGPT

The new dialogue format makes it possible for ChatGPT to reject inappropriate requests, challenge incorrect premises, answer follow-up questions, and admit its mistakes — things that other language models can’t do.

It also produces fewer harmful and false outputs thanks to being trained using Reinforcement Learning from Human Feedback (RLHF). This means human AI writers trained ChatGPT by rating and providing feedback on its performance.

9 Future ChatGPT Products

Now that we know how ChatGPT differs from similar language models, we can better understand how ChatGPT products will differ from existing AI products on the market.

Here are nine we believe will be developed soon.

1) AI Tutors

ChatGTP has already been used as a virtual teacher that explains code as users write it in real time.

Children in Africa learn how to code using Replit and ChatGPT — Source

This shouldn’t surprise us, considering that a previously developed AI model, OpenAI Codex, can already explain code in natural language and act as a tutor.

However, ChatGPT-powered tutors will feel more like real human tutors:

They’ll allow users to ask follow-up questions.
They’ll generate personalized explanations tailored to users’ needs and knowledge levels. This will be a huge advancement considering that current code explanations offered by Codex are basic and not customized for specific users.
They’ll teach more subjects. While Codex can only teach code, ChatGPT could potentially teach an infinite number of subjects.

2) Conversational Search Engines

ChatGPT is already nicknamed “Google killer.” Although we don’t see it replacing Google anytime soon, it could give rise to new, more conversational search engines.

Unlike Google, ChatGPT-powered search engines could give users more specific “results” that directly answer their questions, instead of just giving them resources they could research for answers.

ChatGPT giving a straightforward answer to a question. (Source)

For now, we may have to settle for extensions displaying ChatGPT’s answers alongside Google search results. One, called ChatGPT for Google, is already publicly available.

ChatGPT’s answer to the question “what is AI” displayed alongside Google search results

3) Automated Website Builders

Since ChatGPT can write HTML and CSS, it could soon be integrated with IDEs like Replit and allow users to quickly build websites.

Replit has already demonstrated how users can do this right now, and in under 10 minutes. The entire process can be summed up in four steps:

Prompt. Give ChatGPT a natural language text prompt.
Copy. Copy the code generated by ChatGPT.
Run. Paste the code into Replit to run it immediately.
Repeat. Repeat the steps until you’re satisfied with your results.

See how this works in action:

Replit ⠕ @Replit

If we want a new section added, we ask for the new HTML element. If we want a visual upgrade, we ask for the updated CSS. Prompt. Paste. Run. Repeat.

In the future, we expect Replit and other IDEs to integrate ChatGPT, allowing users to write and run code without going back and forth between different platforms.

4) AI Writing Assistants

ChatGPT can already produce academic papers, grant proposals, marketing copy and content, and even creative written works such as poems, movie scripts, and short stories.

A grant proposal generated by ChatGPT — Source

It can also rewrite existing works in the style of another author or adapt them to new contexts.

John Lennon’s Imagine rewritten by ChatGPT so that it talks about the life of a female startup founder — Source

Admittedly, there are already a number of AI writing assistants on the market, many of which are powered by GPT-3. But we believe that ChatGPT-powered assistants will differ in three important ways:

Users will be able to request iterations more easily. They’ll be able to simply follow up on their previous instructions by continuing the dialogue.
There’ll be fewer limitations. Users should be able to make more specific and diverse requests, from rewriting existing songs to writing sales copy for a specific audience.
The end result will be more refined. Thanks to the two points we mentioned above, as well as the fact that ChatGPT can admit its mistakes, we expect future AI-generated texts to be of higher quality than current ones.

5) Personalized Alert Tools

Even non-technical users can write code using ChatGPT. Among other things, this allows them to create alerts triggered by specific events, such as the price of Bitcoin going up.

A ChatGPT-generated code for alert reminders when the price of Bitcoin goes up by 5% — Source

Thanks to this, we see ChatGPT integrated with schedule, reminder, and alert tools in the future. ChatGPT would help users create code for very specific alerts, while the tools would allow them to implement them.

6) Personal AI Assistants

We expect ChatGPT to soon get integrated with smartphone AI assistants and home automation systems. As a result, we’d get our own personal assistants who could follow complex commands and chat with us in real time.

For now, ChatGPT is limited to writing code that can control AI assistants and smart home devices.

A ChatGPT-generated YAML code for a home assistant automation — Source

Once native integrations roll out, we should be able to add new instructions and expand our devices’ capabilities by giving simple voice commands. The code could be written and implemented automatically.

This would also change the way we interact with our devices. It may allow us to have real conversations instead of just giving basic voice commands.

7) Conversational Art Generators

AI art generators made creating images easier than ever, but some users struggle with writing prompts that would generate images as they envision them. That’s where ChatGPT comes in.

It could act as a middleman and turn simple text or voice commands into prompts that AI art generators can understand.

Chat-GPT generated AI art prompts can create beautiful images — Source

This would let us create suitable images more quickly and without worrying about how we phrase our instructions.

As a side note, it seems like some have already integrated ChatGPT with Stable Diffusion, but no official app has been released yet.

Altryne - targum.video @altryne

My @OpenAI #chatGPT is not like your chatGPT. Mine can... draw 🤪 (using @StabilityAI API or Dalle) Oh and it runs via Telegram, so I can ask it on the go. WIP of course, prompt engineering needs some work!

8) Code Debuggers

ChatGPT can not only debug code but also fix it and explain the changes it made.

ChatGPT debugging code and explaining the fixes — Source

With that in mind, we expect ChatGPT to get integrated with code editors and writers and debug code as users write it.

Unlike Codex-based Copilot, ChatGPT code editors wouldn't require users to enter robotic prompts to fix their code or get an explanation. They’ll be able to pose simple questions or give commands in a conversational way.

As Replit’s CEO put it, “prompt wrangling” will become a relic of the past.

9) Automated Project Management Tools

Since ChatGPT is capable of outlining processes and breaking them down into smaller steps, we believe it has huge potential in the project management sphere.

ChatGPT breaking down the SEO audit process into ten steps — Prompt: “Give me a numbered list of the steps I'd need to take to do an SEO audit of a product page”

If integrated with project management tools, ChatGPT could automatically create tasks and set goals for teams.

Are We Ready For An Influx Of ChatGPT Products?

The list of possible ChatGPT products doesn’t end here. The technology could be applied to many more, if not all industries, and that could be devastating for the future of white-collar jobs.

An additional problem with ChatGPT is that it can bypass the rules set by OpenAI. For example, users have tricked it into accepting inappropriate requests and harmful instructions, despite being “taught” not to do so.

ChatGPT giving ideas on how to bully John Doe — Source

Of course, ChatGPT can be used for many other harmful purposes, such as writing phishing emails and malware.

All this begs the question of whether we’re ready for an influx of ChatGPT products. We don’t know how to regulate its use nor how to save white-collar jobs from going extinct — and, possibly, millions of people from losing their jobs to AI in the near future.

However, it’s not likely that this will stop companies from developing ChatGPT products. So, perhaps the best thing we can do right now is to learn how to apply ChatGPT in our jobs and everyday life.

Our future posts should help you with that, so stay tuned.

Subscribe to get full access to the newsletter and never miss an update.

Here is 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.

Ankur’s Newsletter

Nine ChatGPT Products We Can Expect Soon