Andrew Hsu

Andrew Hsu is the co-founder and CTO of Speak, an AI language-learning company. His work on live tutoring centers on a teaching problem: giving learners room to hesitate, ask questions and practice in their own words while keeping a lesson moving toward its instructional goal. In September 2026, he announced Live Tutor Lessons powered by OpenAI’s GPT-Live-1, beginning with a limited rollout for English and Spanish learners.[2]

Hsu described a consequential finding from user interviews and live testing: many learners did not naturally interrupt a tutor. Speak’s team responded by building in understanding checks and invitations to think and answer in the learner’s own words. Rather than requiring learners to initiate every exchange, the lessons actively create opportunities to participate.[2]

In his September 10, 2026 article, Hsu explained the division of responsibility: Speak defines what learners should practice and accomplish, while GPT-Live-1 adapts the conversation to their responses. A lesson about work, for example, can let someone practice describing their own job and responsibilities. Unexpected questions and simpler explanations remain possible within that instructional direction.[2]

Tutor context changed conversational behavior in Speak’s early evaluation, Hsu reported. Configured as a generic assistant, the model interrupted 27.6% of learner thinking pauses; given the role and context of a language tutor, it interrupted 13.6%. The comparison concerned when the model spoke during pauses, rather than an improvement in students’ learning outcomes.[2]

Hsu also described an end-to-end evaluation using a simulated learner that answered questions and interrupted lessons. Across 27 complete sessions, GPT-Live-1 delivered 476 of 477 authored lesson lines, including in lessons of up to 32 steps. This tested whether conversational flexibility could coexist with delivery of the planned instructional material.[2]

Hsu’s teaching philosophy also favors selective correction. “Correcting every imperfection is not good teaching,” he wrote, arguing that an unnecessary correction can disrupt clarification or successful communication and weaken a learner’s confidence. He acknowledged that the model still missed some errors and sometimes offered stylistic feedback when the learner’s language was already valid.[2]

Before this live-tutor work, Hsu studied neuroscience and began building a startup through the Thiel Fellowship. That earlier history also includes his meeting with future Speak co-founder Connor Zwick.[4][3]

Hsu grew up in Seattle, entered the University of Washington at 12 and earned degrees in neurobiology, biochemistry and chemistry. He began neuroscience doctoral study at Stanford at 16 and left at 19 to join the Thiel Fellowship, where he met Zwick.[3]

A 2012 Fast Company profile dates that transition to 2011: Hsu left Stanford as a fourth-year PhD candidate, became a Thiel Fellow and started working on Airy Labs.[4]