The 4 Supercategories of Generative AI

Hi there,

The recent OpenAI Dev Day really set the stage for what's possible with AI.

But what's becoming clear is that the most impactful applications of GenAI aren't necessarily new challenges; they're familiar problems being solved more efficiently than ever before.

In fact, all of the successful GenAI use cases I've seen fall into a set of 4 distinct problem domains. Today, I'm going to dive deeper into them. Ready for the rundown?

Let’s go!

Most GenAI Use Cases Aren't New, They're Newly Possible

Today's newsletter is inspired by a recent post from ex-Google's Valliappa Lakshmanan. He notes that there are essentially just four good use cases for Large Language Models (LLMs). While I see his point, I think it's more about recognizing patterns in the domains where GenAI thrives, rather than limiting it to a certain number of use cases. Instead, I'd like to think of four "supercategories" of use cases.

In the fast-moving world of AI, it's important to understand that GenAI's greatest achievements aren't necessarily in inventing new solutions to new problems, but in improving our approach to old ones.

Consider chatbots. Automated customer support bots have been around for over a decade. But GenAI has transformed them, making interactions almost indistinguishably human and integrating a vast amount of knowledge. This advancement makes traditional FAQs and inflexible chatbot scripts to a good degree obsolete.

The innovation is not in the what, but in the how.

With this perspective, let's look into the areas where GenAI shines.

Supercategory 1: Summarization and Expansion

LLMs are really good at making existing things shorter or longer. What sounds pretty mundane is actually a key skill in our information age, where the average knowledge worker must process the equivalent of 174 newspapers of information every day.

With the help of Generative AI, we can easily distill volumes of information into short summaries, or go vice versa and turn loose ideas into well-structured, full-bodied content.

Let’s take a look at both concepts:

Summarization: GenAI can easily boil down large documents to their essence. It cuts through the fluff and brings out the essentials in a fraction of the time it would take us to do it manually. As context windows increase, this capability is likely to become even more powerful over time.

Expansion: Conversely, with a burst of an idea, GenAI can act as a creative partner, drafting outlines and filling pages with content - examples, arguments, comparisons, etc. It's like having a brainstorming buddy who's always on the same page with you - literally.

Use case examples in business:

• Create a marketing strategy from a simple product pitch.

• Generate a series of social media posts from a single topic.

• Boil down dense website content to powerful SEO keywords.

• Summarize long reports or processes into key ideas or steps.

Like Akiff here, who asked ChatGPT to summarize his inputs on the complex process of how the supply chain works in the pharmaceutical industry:

But beware: Even the most compelling sounding text summary can actually be incorrect, omitting important facts or making up new ones.

That's why - especially in the beginning - strong human oversight and double-checking of AI results is critical, underscoring the need to start with augmented AI use cases.

Supercategory 2: Information Retrieval

Another category where Generative AI - and especially LLMs - excel is in understanding the semantics and nuances of a user's intent. For example, when I ask ChatGPT to "jot down some minutes for me," it recognizes that I'm talking about meeting notes, not time.

This precision can be a game-changer for two key areas:

Search: With LLMs, you're no longer limited to fuzzy keyword searches. Instead, you can search by semantics. Using a technique called vector embeddings (translating the meaning of the term you're looking for into a numeric array) and an architecture called Retrieval Augmented Generation, LLMs lets you build the most powerful search engine for your proprietary documents.

Knowledge Retrieval: Because of their huge amount of training data, LLMs provide access to a huge amount of knowledge right out of the box. It is like having a cheat sheet for almost everything. Especially when it comes to proven business frameworks like SWOT for business strategy, Jobs-to-be-done for product management, or BANT for sales calls, LLMs can walk you through these frameworks step-by-step to solve the task at hand.

Use case examples in business:

• Improved Product Search: Imagine an online store where you can search for "comfortable summer shoes" and the results are just showing you the perfect pair.

• Customer Support Chatbots: Chatbots can pull specific answers from a vast knowledge pool, making customer interactions much smoother and smarter. Here’s another article on this:

In short, LLMs help you figure out what you want, even when you're not quite sure how to ask for it. This doesn't just make new use cases possible — it makes them feasible.

Supercategory 3: Creative Assets

The third supercategory is generating creative (digital) assets. With GenAI, ideas can take form and color, motion and sound. The applications are as limitless as creativity itself:

• Turn text into images (e.g. Dall-E)

• Expand images using outpaint (e.g. Midjourney)

• Manipulate image areas using text (e.g. Adobe Firefly)

• Turn text into voice (e.g. OpenAI TTS)

• Turn speech into text (e.g. OpenAI Whisper)

• …

This AI-enabled "multimodal magic" allows you to traverse between different types of modalities with an ease and speed that was previously impossible. This unlocks a plethora of "old" but now even more exciting business use cases:

Use case examples in business:

• Branding Material: Generate variations of logos, banners, and other branding elements to fit a given context (e.g. different ad sizes) without a design team.

• Custom Stock Images: No need for generic stock photos when you can create custom imagery that perfectly suits your content.

• Product Visualization: Before the first prototype is ever built, GenAI can provide a realistic visualization of your product, saving time and resources in development. Here’s a strong example of this:

With digital asset creation and AI, the barrier between idea and execution just got a lot thinner.

Supercategory 4: Coding

The final supercategory, and by some accounts the most pivotal, is coding.

AI-powered coding tools like GitHub Copilot promise to speed up developer productivity by over 50% - amongst other benefits. It's a game-changer, not only for improving the skills of experienced coders, but also for enabling novices to quickly build complex programs.

Using AI, people with minimal coding experience can now contribute in areas once reserved for the few with highly specialized skills.

This is a huge leap in democratizing the development process, turning creators into coders.

(And yes, let's skip the sci-fi scenario of AI coding itself into autonomy for now).

Here's how AI-powered coding can drive real business results today, even beyond traditional developer roles:

Business use case examples:

• Data Analytics: Answer complex queries such as “What percentage of customers is driving 80% of my revenue” by having an LLM write the code required to uncover these insights.

• Data Access: Valuable data is often locked away in data warehouses or buried in unstructured PDFs and text files, out of reach for less tech-savvy people. AI-powered coding capabilities are changing that, providing access to insights that once required a gatekeeper.

GenAI in coding isn’t just about making existing developers more efficient — it's about expanding the definition of who can be a developer.

And especially with regards to coding, we’re just about to see the beginning:

Conclusion

GenAI is a powerful tool that can help you innovate and grow. Don't wait for the perfect moment or the next big breakthrough.

The tools we have today are already powerful enough to move your business significantly forward.

Start with your business problems, and then see how AI can help you solve them better - as a catalyst for growth. That's the best and most sustainable way to drive digital transformation with AI.

I hope you enjoyed today's newsletter and now have more clarity on which GenAI use cases to pursue.

If you're not sure where to start, just reply to this email and let's talk.

See you next Friday,

Tobias