CHAPTERS
- 0:00 – 1:00
Tom Yeh’s mission: making AI understandable “by hand”
Tom Yeh introduces himself as a CU Boulder computer science professor and founder of AI by Hand, an initiative aimed at demystifying AI by writing out the math manually. He frames learning as ownership and internalization, not just having correct outputs or credentials.
- •AI by Hand’s goal: open the “black box” by hand-writing the underlying math
- •AI is approachable if you build conceptual understanding
- •Learning is about ownership, not artifacts like certificates
- •Having an answer isn’t the same as understanding
- 1:00 – 2:01
Relearning deep learning from scratch—and sharing the struggle
Yeh describes missing the first wave of deep learning as a student and later needing to catch up quickly as a professor. He explains how drawing and writing became his primary tool for truly understanding models and algorithms, and why that resonated with others.
- •Background in traditional ML (e.g., SVMs) before the deep learning boom
- •Had to relearn deep learning after becoming a professor
- •Handwriting/drawing as a mechanism for genuine comprehension
- •AI by Hand emerged from sharing his learning process publicly
- 2:01 – 3:31
Why teaching “slow” works: blackboard coding and human-speed learning
He recounts switching an intro C++ course to full blackboard teaching after feedback that he moved too fast. The format naturally enforces a pace aligned with human processing and improves student focus and attention.
- •Blackboard teaching limits instructor pace to writing speed
- •Students can only track what’s written—reduces cognitive overload
- •Hand-copying notes keeps students off distractions (e.g., phones/Instagram)
- •“By hand” reconnects learning to human constraints and attention
- 3:31 – 4:02
Redefining learning in the AI era: ownership over instant answers
Yeh challenges the idea that access to AI-generated solutions equals learning. He argues that value and retention are tied to time, effort, and personal struggle—not the immediacy of an answer.
- •AI can supply answers instantly, but that doesn’t build understanding
- •Credentials can be purchased; comprehension must be earned
- •Knowledge is valued in proportion to time invested acquiring it
- •Learners must define what learning means to them
- 4:02 – 5:33
Evergreen foundations vs. fleeting tools: the matrix multiplication lesson
Using matrix multiplication as an example, Yeh shows how core concepts recur across waves of technology—from graphics to big data, machine learning, AI, and even quantum computing. He contrasts these durable foundations with fast-changing, hype-driven tools.
- •Core ideas (e.g., linear algebra) reappear across major tech shifts
- •Tool trends come and go; foundations persist
- •Skepticism about chasing short-lived products and hype cycles
- •Transformers likely persist longer than many named tools/models
- 5:33 – 6:34
The palace that burned: skills rebuilt on solid foundations
Yeh tells a story about a Korean palace that was rebuilt after burning down because its rock foundation remained. He maps this to careers: if you invest in fundamentals, you can rebuild and adapt when technologies shift.
- •Gyeongbokgung palace rebuilt using the original foundation after destruction
- •Tech changes repeatedly; foundations enable rebuilding
- •Focusing only on surface tools forces constant restart
- •A strong base makes new tools easier to learn and integrate
- 6:34 – 8:05
Your enduring advantage: the meta-skill of learning hard things
He encourages viewers to recognize skills developed through long practice—piano, soccer, chess—as evidence of an enduring ability to learn. That learning capacity becomes transferable leverage when navigating new AI tools and domains.
- •Past skill-building proves you can acquire difficult abilities
- •What’s durable is the learning process, not the current tool
- •Skipping a trendy tool is fine; abandoning skill-building habits isn’t
- •Identity and long-term capability come from sustained practice
- 8:05 – 9:36
What education should optimize for: willingness, effort, and challenge-taking
Yeh explains that students may forget specific equations, but what matters is the demonstrated willingness to engage with complexity and build foundations. The differentiator is effort and persistence, not short-term memorization.
- •Long-term retention of details is less important than learning capacity
- •“Opening the black box” signals courage and intellectual commitment
- •Foundation-learners demonstrate time/effort investment
- •Avoiding challenge signals lack of willingness to do hard learning tasks
- 9:36 – 10:36
Cheating tools are symptoms: incentives drive shortcuts
He reflects on combating cheating via Chegg, only to see AI replace it as the new shortcut. This leads to a broader critique: the underlying incentive structures push students toward cheating, regardless of the specific tool available.
- •Educators spent effort countering solution sharing and cheating
- •Chegg declined, but AI replaced it—problem persisted
- •Cheating is driven by root incentives, not just platforms/tools
- •Fixing education requires addressing why students feel compelled to cheat
- 10:36 – 11:37
The ‘AI-native’ trap: hire for fundamentals, not buzzwords
Yeh argues that organizations should prioritize hiring for work ethic, problem-solving, and teamwork. People with these traits will naturally adopt AI effectively; forcing “AI-native” expectations often signals misaligned hiring priorities.
- •Employers ultimately value work ethic, problem-solving, communication
- •Strong problem solvers will learn AI without being pushed
- •Team players will use AI to collaborate better by default
- •Overemphasis on “AI-native” can reflect hiring the wrong fundamentals
- 11:37 – 12:37
AI can’t change people—but people can change AI
He closes with a boundary: AI tools won’t transform character traits like respect, integrity, or collaboration. Human agency and values must lead; only then can people shape AI systems and uses toward better outcomes.
- •AI won’t make someone a team player or ethically grounded
- •Character and habits remain human responsibilities
- •Use AI as an amplifier of good fundamentals, not a replacement for them
- •People can reshape AI’s impact through choices and design
