Leya & Large Language Models

January 15, 2024

Welcome to Leya's blog – hello world 👋!

In this digital age, where we're bombarded with information, the real superhero power we're all after is learning. We're talking about the kind of learning that doesn't just throw facts at you, but actually gets you. That's where we at Leya AI swoop in. Our mission? To make personal coaching as common as your morning coffee, available to everyone, everywhere!

But here's the techy bit – our secret sauce is Large Language Models (LLMs). They're the backbone of our mission, enabling us to tailor personal development experiences like never before. Stick around as we dive into how we're using LLMs to revolutionise learning and growth!

Current Focus

In our mission to build personal AI coaches, our current focus is on crafting an exceptional AI-English Language Tutor. By partnering and integrating the strategies of top human tutors into our AI-Tutor, we've harnessed LLMs to offer tailored, interactive learning experiences that adapt to individual student interests.

Our Approach

The introduction of platforms like ChatGPT has completely changed the perception of AI capabilities. However, harnessing these technologies for educational purposes is not without challenges. A good thought experiment, that helps to judge whether an LLM could do some task, is to imagine orchestrating an 'army of 9-year olds' – an army full of potential but requiring a lot of guidance to achieve any meaningful outcomes.

I would lie if I would say that we have it all figured out - quite the contrary, LLMs are a hot topic in research, and utilising them to their full potential is a state-of-the-art problem. Internally we’ve built lessons in Leya with 3 core steps:

1. Prompt Development

Crafting engaging and effective prompts that do what we expect (frequently enough 😅)

2. Conversation Execution

Ensuring dynamic and interactive AI conversations that mimic a real tutor's approach. This topic will require its own blog post as this is where a lot of the application's complexity is concentrated.

3. Collecting Feedback

We collect both explicit and implicit feedback wherever possible.

💡 Working with LLMs has introduced unique development challenges. For example we had an interesting incident which involved rolling out a prompt improvement that led to the LLM unexpectedly withholding feedback from almost all of our students.

Some debugging later we found the problem and issued a bugfix which was as simple as altering a single word (from below to above), but it underscored the nuanced nature of LLM-driven solutions.

The Future

As LLM technology rapidly evolves, so do our opportunities at Leya. We are actively exploring advanced tooling and methodologies to enhance our AI coaches. The potential to deliver more personalised and effective coaching, previously unthinkable, is now within our reach. We just need to become better generals of the 9-year old army 🪖 Stay tuned as we navigate this exciting frontier!

Related posts

Leya AI Secures €1M Pre-Seed to Transform Global Education

February 6, 2024