Covers 1:09:50–1:18:32. What Martin teaches at Cambridge, and how AI is disrupting assessment.

What he teaches

  • Current load: concurrent and distributed systems (undergraduate), cryptographic protocol engineering (masters), a security seminar, and the undergraduate operating systems course. 1:09:50
  • The distributed systems course is freely available on YouTube: more theoretical than the book, focused on algorithms and on convincing yourself they behave correctly under the standard assumptions (nodes crash, communication unreliable, clocks wrong). Eight lectures; one goes through the entire Raft consensus algorithm — chosen because it illustrates the challenges and shows consensus is solvable despite subtle, bug-prone algorithms. 1:10:25 1:10:57 1:11:20
  • Teaching style: slides annotated by hand on an iPad; mostly theoretical pen-and-paper (the Cambridge way), except the cryptography course, which is hands-on — students implement elliptic curves from scratch. 1:11:49 1:12:08

How AI is changing education

  • Pre-AI, CS teaching changed very slowly — Cambridge (800+ years old) focuses on fundamentals (many from the 1930s, e.g. lambda calculus) over fashionable fads. 1:12:38 1:13:04
  • AI totally changed coursework assessment: banning it is unenforceable and counterproductive (students should learn to use new technology), but usage must support learning rather than undermine it. Many students aren't mature enough to draw that line themselves, so guardrails are needed; perceived fairness of assessment is at stake — if some students get top marks without work, trust in the whole system erodes. 1:13:36 1:14:09 1:14:27
  • Concrete response so far: a first-year bootcamp on basic software engineering skills — version control, unit testing, generative AI. Assessment policy is still unresolved: "we don't really have good answers yet". 1:14:52 1:15:17

The key distinction: outcome vs. process

  • Industry's desired outcome is a working product — if AI gets an equivalent result faster, use it. In education the artifact (the essay) is not the point; the thought process and learning are. So education must preserve the thinking even when tools could skip it. 1:15:40 1:16:07
  • Referenced: a recent Anthropic study of junior engineers where the AI-using group showed little to no learning while the non-AI group actually learned. Martin: methods might be quibbled with, but the principle holds — learning requires struggling a bit. AI is good for getting unblocked on technicalities; bad when it replaces grappling with the core ideas. 1:16:21 1:16:42 1:17:03

Bridging the two worlds

  • Industry and academia often disrespect each other ("that's academic, nothing to do with the real world" vs. "that's just routine engineering") — both miss out. Martin sees it as one of his goals to build respect in both directions: bring research insights into practice, and let real-world problems inform research. 1:17:28 1:17:47