In the current era of AI, it seems that much of the challenge lies not just in harnessing its power but in making it accessible to users. It's akin to having access to an extensive library without a clear guide on where to begin. While there’s an outsized focus on model power and learning how to work with a “raw” GPT, I'm particularly interested in how we can create user-friendly interfaces that simplify entry into various fields.

The great Tim Urban uses a neat analogy^[1] to describe learning: imagine knowledge as a tree. Without a solid trunk - the basic understanding - any new info doesn't stick. It's like adding branches to a tree that can't support them. This is especially true for AI; without a straightforward starting point, its advanced features are virtually inaccessible.

The main issue with today's AI is not its capabilities but its complexity. Navigating through countless ChatGPT prompts or mastering "prompt engineering" demands a certain level of expertise. It reminds me of the early days of learning to code, where knowing how to search for your problem is half the challenge. To effectively utilize AI with today’s tools, understanding how to frame your questions is crucial.

This complexity opens doors for businesses to develop tools that customize AI for specific areas. Picture a platform that introduces you to a new topic with AI-generated content to build your foundational knowledge, coupled with user-friendly features for further exploration and learning. This approach could incorporate specific terminologies into the user experience, providing guidelines to ensure users derive maximum benefit.

Take music discovery, for example. Spotify’s approach to this problem is is passive — it tracks what you listen to and gradually introduces similar music to your playlists. This is a great experience, but it’s both opaque (what aspects of the music drove your recommendations?) and can trap you in a loop of the same old tunes. What if you want to proactively explore new music? Enter AI: imagine being able to input a song or artist you’re interested in, have the model break down what makes the music tick, and then recommend new stuff based on the qualities you’re most interested in. It’s a much more intentional way to discover music that suites your taste.

Cooking is another domain that could benefit from this approach. Say you have a favorite flavor or cuisine; AI can sum up the key elements of that cuisine for you. Armed with that information, you can better navigate your culinary journey by asking for similar flavors or for tips on expanding your repertoire in that cuisine. This also feels like a natural extension of the “cooking without recipes” angle — recently, I asked ChatGPT for a basic template for a Chinese-style stir fry sauce. It gave me a (tasty) template with suggestions for how I can customize it depending on what I have on hand, but it took a lot of trial and error in my prompts to figure out how to ask the right question.

Creating intuitive AI tools for specific domains presents a valuable opportunity for product innovation. Starting with a basic AI-driven introduction to a topic, these tools can then distinguish themselves through a tailored experience that makes it easier for users to build on that knowledge. This strategy benefits end-users by transforming the vast ocean of AI information into a manageable stream, guiding them toward the knowledge they seek. Such an approach not only simplifies learning but also personalizes the journey, allowing individuals to pursue their interests in a way that's most meaningful to them.