Speakers
Description
Do students even need real teachers anymore? That is a question floating around Silicon Valley in light of recent advancements in AI. Are they caught up in the hype, or is there a foreseeable path to AI teachers?
The present state of LLMs leaves a lot to be desired for advanced applications designed to embody intelligent language teachers. We cannot simply prompt our way to a solution with existing models. A broad-spectrum approach combining technology across machine learning disciplines is imperative in building rich experiences for students. Essential for this mission is an ocean of quality data - LLMs for language learning require specific data that captures the language learning experience.
ImmersiveAI is a startup attempting to develop an AI language tutor. In doing so, our company is pushing the limits of AI and assessing the learning outcomes for 20,000 language learners. ImmersiveAI has developed key insights into exactly how AI needs to improve before it can replace a tutor and where it already excels. Currently, AI enables the mass proliferation of learning techniques centered around comprehensible input and conversational interaction, as demonstrated in our application, Crosstalk. Advanced text-to-speech and prompt-imbued personalities keep students engaged, whereas earlier chatbots felt like talking to a dull robot.
Personalized level-matching tailors the learning experience to each student, thereby facilitating greater understanding. Contextual awareness enables improved error correction and translation. Even though intelligence capable of replacing human tutors does not yet exist, student outcomes at this early stage are promising.
Language learning is transformative for the individual and society at large. Very few students have access to a private tutor, and many students rarely have the opportunity to practice with anyone at all. If everyone had a private language tutor in their pocket available 24/7/365, what would the world look like? How do we get there?
| Keywords | ML Engineering, Immersion, Comprehensible Input |
|---|
