Launch HN: Miyagi (YC W25) turns YouTube videos into online, interactive courses

Overall Reception and Concept

• Many commenters find the idea “magical,” intuitive in hindsight, and a natural fit for LLMs (auto‑generating quizzes, tutors, and structure around existing videos).
• Several educators and edtech practitioners say this solves a real pain point: turning scattered YouTube learning into something structured and interactive.
• Others are more skeptical, noting users can already feed transcripts into general-purpose LLMs and asking what value Miyagi adds beyond convenience and UI.

Content Quality, Pedagogy, and Trust

• Big concern: how to ensure generated courses aren’t “teaching nonsense,” especially for learners who can’t detect subtle errors.
• For “official” courses, creators review and edit content; this is currently treated as the main quality signal.
• For user-generated courses, there is no systematic human review; future plans include user feedback on questions, learning paths, and possibly more formal evaluation.
• Edtech professionals raise questions about deeper methodologies (knowledge tracing, skills taxonomies, learning outcomes, retention), which are not yet clearly implemented.
• Some argue assessments are overemphasized; for self-directed adult learners, quizzes may be more “demo” than true value unless backed by tracking, validation, and credentials.

Experience, Features, and UX Feedback

• Positive feedback on the general interface and idea of integrated tutor, quizzes, and flashcards, but multiple users report bugs, long generation times, and login issues.
• Suggestions:
  • Clean up long lecture lists and better sectioning.
  • Integrate bottom tools into a single, more agent-like tutor.
  • More gamification, cohorts/social learning, and real-world artifact-based tasks, not just trivia.
  • Smarter “watch” links that jump to the relevant video segment.
• Debate over giving the AI tutor a “persona”: some like the idea; others find overly humanized AI off-putting.

Use Cases and Scope

• Interest in:
  • Language learning (with repetition and grammar focus).
  • Child/elementary content, though the current tutor safety is not trusted for unsupervised young kids.
  • Poker, chess, and other domains that may need custom tooling (solvers, interactive boards).
  • Niche “how-to” content (e.g., home improvement), including the potential use of comments as errata.
• Currently relies mostly on transcripts; no video-frame understanding yet, though that’s on the roadmap. Supports PDFs, slides, and other file types as inputs.

Copyright, Ethics, and Creator Relations

• This is the most contentious theme.
• Miyagi says:
  • Any monetized course includes revenue sharing with creators, with some signed partnerships.
  • Embedded videos are used, and creators can request takedowns.
  • Non-partner content can still be summarized and augmented, with opt-out rather than opt-in.
• Critics argue:
  • Opt-out with ambiguous revenue terms is ethically weak; should be explicit opt-in and generous sharing since creators supply most of the value.
  • Even if embedding is allowed, generating derivative educational products without consent feels exploitative to many.
  • “Extra views” are not sufficient compensation if a platform monetizes AI-based derivatives.
• Comparisons are drawn to:
  • YouTube’s own AI summaries, accused of reducing watch time while monetizing summaries without adequate creator compensation.
  • Earlier controversies where companies enrolled creators into monetized systems without clear opt-in, which produced severe backlash.
• Legal status (derivative work vs. fair use) is left unresolved in the thread; multiple commenters flag this as a major long-term risk and perception issue.

Platform Risk and Future Direction

• Some discuss dependence on YouTube APIs and subtitles; founders state they already support direct uploads and could expand to other subtitle sources.
• Commenters from traditional edtech note this idea has been floated before but ran into pedagogy, licensing, and “no-ads” constraints at established platforms, hinting that a startup can move faster but must still address those same issues over time.