Revolutionizing Language Learning with AI: Building Duolingo Max

Revolutionizing Language Learning with AI: Building Duolingo Max

Table of Contents:

I. Introduction II. The Duolingo Story III. Embracing AI in Language Learning IV. Building Duolingo Max V. Lessons Learned in Building Duolingo Max VI. The Role of Prompts in Generative AI VII. Creating Compelling Role-Play Conversations VIII. The Future of Generative AI in Education IX. Conclusion

Article:

Introduction Generative AI has the potential to revolutionize the way we approach education products. At Duolingo, we have been exploring the possibilities of AI in language learning for years, and with the release of gpt4, we saw an opportunity to take our products to the next level. In this article, we will share our experience building Duolingo Max, our newest and highest-tier subscription powered by open ai's gpt4. We will discuss the lessons we learned along the way and provide insights into how we created compelling role-play conversations that showcase the unique personalities of our characters.

The Duolingo Story Duolingo is the worldwide leader in language education, with over 500 million learners studying 100+ courses daily. Our mission is to develop the best education in the world and make it universally available. To achieve this, all of our content on Duolingo is completely free, and learners can make it through an entire course without ever having to pay a cent. However, monetization at Duolingo exists for the very important purpose of fueling that mission. Our super Duolingo and now Duolingo Max subscription tiers allow us to hire more talented people, teach more courses, and ultimately teach those courses better to our millions of learners.

Embracing AI in Language Learning AI has always been an integral part of our strategy at Duolingo, allowing us to improve our teaching methods in scalable and impactful ways. However, advancements in AI and specifically generative AI have forced us to approach the technology differently and start thinking about how we can build products and experiences with an AI-first approach rather than just using it to supplement our existing product. With the release of gpt4, we saw an opportunity to create a product that could replicate tutor-like quality instructions, something we had been striving for but thought was still four to five years away.

Building Duolingo Max Building Duolingo Max was a small team effort of about 10 dedicated engineers, designers, animators, and data scientists. We had just about five months to build two new features completely from the ground up, an entire subscription tier to house the product, and ensure that it was something that would not just completely run us out of cash. We identified four learner needs that we thought gpt4 might be able to help solve: better understanding of grammar and language rules, improved speaking and pronunciation skills, feeling more prepared for real-world conversations, and having a better understanding of their progress in our courses.

Lessons Learned in Building Duolingo Max We learned five key lessons in building Duolingo Max. First, we haven't won yet, and it's always important to keep staying on top of recent trends and continuing to innovate to stay ahead of competition or to stay ahead of new technologies. Second, find your product's unique edge and marry technical possibility with core user needs to find that edge. Third, start with the prompt, don't just jump straight into high-fidelity mocks or wireframes before you actually validate what AI will look like in practice for you. Fourth, generative AI is a tool for subject matter experts, not a replacement. Fifth, don't forget what makes generative AI magic and start by building a really big body of examples of what kind of outputs you might actually want to see from your prompts.

The Role of Prompts in Generative AI The prompt is the foundation of any generative AI product, and it's important to start with the prompt and not just jump straight into high-fidelity mocks or wireframes. We learned that prompt validation needs to happen earlier in the process than we originally thought, and it's important for everyone on the team to understand the prompt strategy and how that can actually motivate your design UX.

Creating Compelling Role-Play Conversations Role-play conversations are a welcome addition to Duolingo, and learners were excited about how they could use that to work towards their language goals. However, we realized that the main thing that role-play was lacking was the voice quirk and storytelling that makes Duolingo Duolingo. To make these conversations more rewarding and ensure that these character personalities could really shine through, we brought in storytelling masterminds to create a set of principles that defined what a rewarding role-play conversation needed to look like.

The Future of Generative AI in Education Generative AI has the potential to revolutionize the way we approach education products, and we're excited to see what the future holds. We're constantly assessing and iterating our prompts so that our character personalities can shine through even stronger, and we're thinking about a bunch of different ways we can actually make role-plays feel even closer to the experience of a real conversation.

Conclusion Building Duolingo Max was a small team effort that taught us many valuable lessons about the role of prompts in generative AI, finding your product's unique edge, and the importance of subject matter experts in building generative AI products. We're excited to see what the future holds for generative AI in education and can't wait to continue pushing the boundaries of what's possible.