Personalized AI Language Education — with Andrew Hsu, Speak

Andrew Hsu, co-founder of Speak, shares his journey from being a Teal Fellow to launching an AI-driven language education platform. He emphasizes the fellowship's influence on his entrepreneurial spirit and the connections he made, which were pivotal in his career. Hsu's fascination with AI, particularly in speech and language models, led to the creation of Speak, aimed at developing AI language tutors for faster fluency compared to traditional methods.

The company initially focused on custom speech technology before advanced language models like Whisper and ChatGPT emerged. They found success in South Korea, primarily teaching English, and tailored their speech recognition models to enhance understanding for non-native speakers. The app features a low-latency recording loop and a "magic onboarding" feature that personalizes user interactions. Hsu critiques existing language learning apps for neglecting speaking practice and categorizes the evolution of language learning technology into three generations, culminating in AI-powered solutions that emphasize practical conversations.

In 2018, Speak pivoted to focus on the South Korean market, integrating LLMs and Whisper technology to provide direct feedback on language use. The app has become the largest English learning platform in South Korea, with plans for expansion into other Asian markets and the U.S. Hsu discusses the evolution of their B2B initiative, which has grown from a side experiment to a significant focus, and highlights the competition between language learning and real-time AI translation.

Cultural insights reveal the challenges of entering the Korean market, particularly the importance of app localization. Speak has developed a proprietary teaching method and invested in in-house content creation. Hsu notes the pivotal moment in 2022 with the release of OpenAI's Whisper, which validated their predictions about advancements in speech and language models, leading to significant product improvements.

The conversation addresses the high failure rate of language learning and the willingness to pay for effective solutions. Hsu envisions using AI to transform learning across various subjects and emphasizes the need for direct AI content generation to scale language offerings. The development of various agents for language learning, including a tutor agent and a curriculum writing agent, is underway, with a focus on quantifying real-world fluency.

Challenges in conversational skills are acknowledged, and the evaluation of AI-generated courses relies on human oversight due to the complexities of language instruction. Accent training is highlighted as crucial for effective communication, and the marketing strategy promotes private practice to build confidence. The potential for real-world immersion tools and the integration of audio and video in language learning is discussed, along with the need for a voice mode for on-the-go learning.

Hsu reflects on the strategic acquisition of the domain name speak.com to build a global consumer brand aimed at reaching a billion language learners. He expresses optimism about technology's potential to benefit humanity and emphasizes the importance of focusing on user needs in navigating the anxieties surrounding AI advancements. The conversation concludes with gratitude for the discussion.