A generative AI attracted wide attention in the year 2022. With the help of Dall-E, Midjourney, and Stable Diffusion, we learned how to create images. Many blog posts, including the one you're reading now, contain parts that have been edited or even entirely written by AI assistants. Many pieces of software are co-written using GitHub's Copilot and other tools.

What can we anticipate for 2023 given this year's AI explosion? Is it going to be as crazy as this one?

There is undoubtedly a lot of room for surprises, and talk about the next big breakthrough is rife. However, some events appear to be more likely to occur than others. So here's my prediction for how 2023 will pan out.

AI as a feature

While in 2022 startups used AI to create new apps that wanted to replace existing ones, I believe the emphasis will shift to AI powering up existing tools.

Current productivity and creativity suites control the user distribution channels and thus have the competitive advantage over any new startup.

Text suggestions, grammar, styling, and wording corrections based on Open AI GPT-3 are already appearing in many tools, and they will become more common in the coming year. So, if your product is about writing that does not support GPT-3 or similar you will lag behind the competition in 2023.

Notion integrated AI to help users to write better and brainstorm ideas

Image creation AIs will make inroads into both design suites and business presentation suites as a feature. In this case, I believe that open-source models such as Stable Diffusion will be mostly used. We've already seen how Stable Diffusion-based plugins for Photoshop and Figma are created.

Stable Diffusion plugin for Figma to add image assets to design project

Also, Stable Diffusion based image creation will be added to business presentation software like PowerPoint or GSlides. Perhaps we can also see plugins and features in presentation apps that support style transfer, compositing, and simple 3D rendering.

Corporate imagery services and photo stocks will eventually incorporate one or more AI image-creation models directly into their services. AI-generated images are already available for purchase in large quantities on stock photo websites such as Shutterstock, iStock, and Adobe Stock.

Alternatively, users can head up straight to AI-created photo stock like stockai.com which provides 1000s of pre-created photos already and you can create your own unique one by simply inserting a text prompt.

Incumbent photo stocks, however, have a significant competitive advantage over newcomers as they already have a licensed vast pool of imagery to train their own models. It will only be a matter of time before they introduce their own prompt engines for users. And that could be a serious threat to any other stock image creation service out there.

AI as a product

In order for an end-user AI-based product to succeed in 2023 and beyond, it needs to be vertically integrated. It must control and develop its own model that would exceed the open source or rented AI models’ capabilities notably.

Midjourney is one such product that is currently outperforming the competition. Its imagery is far more pleasing and artistic than that of an open-source model such as Stable Diffusion. This gives Midjourney a clear advantage over any other image-creation software on the market.

Same prompt different results: Stable Diffusion on left, Midjourney on right

If new image creation apps based on Stable Diffusion want to succeed next year, they need to start training their own derivative SD models or provide users with the tools to train their own models easily. Also, a lot can be achieved by improving user experience.

It's much more difficult to forecast the future of new text-writing apps.

Many tools are emerging that use Open AI's GPT-3, and it is difficult to see how they will differentiate themselves from a product standpoint.

In that category, such new apps may compete in terms of UX or pricing, but they will struggle with distribution once major office productivity suites have their GPT-3 integrations up and running.

However, tools such as Lex, Quillbot, and others are currently making the life of a writer much easier when compared to writing in Google Docs or Microsoft Word.

These new apps provide immediate assistance in better phrasing, correct grammar, and even ideation.

Writing this blog post on Lex.page

In the code-generation field, the current leader is GitHub Copilot, which is now used by high-profile companies such as Google, Facebook, and Netflix. GitHub Copilot has a clear advantage over any other code-generation tool on the market.

In order to compete with GitHub Copilot, new code-generation tools need to focus on providing a better user experience and integration with existing development workflows.

AI models as a service

The companies that control the development of generative AI models and provide them over APIs will see their growth numbers skyrocket in the coming years. A lot of app developers need access to their underlying models because their core competencies do not allow them to develop their own. Also, the need for speed to market drives many app builders towards pre-existing models.

AI models as a service field will be dominated by the tech industry behemoths like Google, Meta, Microsoft, and others. However new independent companies grown up from research labs, like OpenAI and Stability AI have gained significant traction this year and will probably grow next year as well.

Notable users of OpenAI API

The only threat to proprietary AI-model developers may come from open-source models. One such battle will be between Dall-E and Stable Diffusion next year. Stable Diffusion has grown a huge open-source community that develops tools and iterates on the derivative models. For Dall-E to survive against Stable Diffusion their output should become as good or better than Midjourney quickly.

AI-assisted services

There seems to be a lot of potential in providing AI-assisted services. Creative or business service companies that tap into AI's superpowers will quickly overtake their competitors who have not upped their game yet.

In the creative field, I already see that AI artists are creating their digital art shops on Etsy to sell their images for interior design. I predict we will see a lot more AI imagery appearing in Ads, content marketing, apparel, etc.

An example of AI generated art sold on Etsy

The same trend applies to copywriting. Whether you write copy for content marketing or SEO, those services that use AI-assistant will quickly outperform their competition.

Finally, on the business services side, I predict a significant increase in the number of AI-assisted consulting services. The services will be built around giving advice on AI adoption, AI deployment, and AI-driven sales, marketing, and PR.

Computing power as a service for AI end-users

Finally, I think that within the coming year, tools will start to appear that will make renting GPU power to end users a commodity.

Do you want to make a movie of your bachelor party in the style of Game of Thrones? Sure, no problem! Would you like a holiday slideshow based on Disney’s  Frozen? Doable.

To get started, simply train your own stable diffusion-based model.

However, doing so places a significant load on the GPUs, and the entire process will take hours to complete unless you have a high-end gaming laptop. Few people are that committed to the job.

In 2022, people solved it by borrowing computing power from Google using Colab notebooks, but whoever develops an app that extends my 4y old Mac GPU seamlessly to run a training session for Stable Diffusion locally will get my money.

Running a Gradio web UI for Stable Diffusion on Google colab

Also, I believe 2023 will be the last year in which we pay attention to how many pixels and what lenses your new iPhone will have.

It won't matter much longer because users are flocking to AI image creation, and the phone camera will soon be obsolete.

The GPU processing power on whatever device you want to use to create your own images and videos using creative AI models will be more important.

Final thoughts on generative AI in 2023

In conclusion, the next year will see the barriers to entry for generative AI applications falling rapidly. As a result, we will see an explosion of new AI-based products and services.

The companies that will succeed will be those that provide either better models or better user experience than their competition.

As an end-user, you will also have a lot of new attractive options to incorporate AI-based superpowers into your daily life next year.

You can easily improve 10x or 100x at art, writing, coding, or whatever your personal calling is. The only issue is that you won't be able to keep track of everything that happens in the coming months. The speed of development is insane in the generative AI field. In order to not burn out quickly, remember to take pauses regularly and enjoy a cup of tea or small walks in not AI-created park.

Cover photo: /imagine prompt: Generative AI in year 2023, Midjourney