YouTube

More than 500,000 AI and machine learning tutorial videos landed on YouTube across 2024 and 2025, pulling over 2.5 billion views a year, with 100 hours of new AI content uploaded every hour and searches for "AI tutorial" or "machine learning course" up 800% year over year. That output exceeds traditional academic production. For many aspiring practitioners the platform has become the primary learning resource, which puts an advertising company in the position of a technical school.

Channels run by Andrej Karpathy, StatQuest, 3Blue1Brown and Sentdex hold millions of subscribers, and Stanford and MIT professors publish there for free at a production quality that rivals paid courses. The same feed carries tutorials built on deprecated libraries, conceptually wrong explanations, and clickbait promising unrealistic results. Views and likes are imperfect quality signals, so learners have to build their own discernment. Audiences compensate by self-organising: curated playlists, Discord and Reddit communities, GitHub "awesome" lists.

Simon Wardley's mapping method uses YouTube as its worked example of a commodity. The video idea and the content sit on the left of the map where differentiation lives; the camera is a product, the platform is a tap you draw from and never rebuild. Think with Google's 2026 budget guide treats the same organisation as a line item, telling marketers to test Shorts and measure post-view conversions. Algorithm dependence, creator burnout and platform risk stay unresolved in both readings.