Lesson 19 showed expertise is built from chunking, acquired through practice. This lesson asks the harder question the power-law-of-practice framing (Lesson 18) glosses over: practice reliably improves performance early on — so why do so many experienced people plateau, sometimes for decades, well short of true expertise?
Terms (standalone):
- Naive/routine practice: simply performing a task repeatedly, at a comfortable, familiar level of difficulty, without deliberately targeting weaknesses or seeking corrective feedback. Reliably produces improvement early on (Lesson 18’s power law), then plateaus — performance stabilizes at “good enough to get by” and stops improving further, often for years, even with continued high volume of repetition.
- Deliberate practice (Ericsson’s term, from research often oversimplified in pop-science as “the 10,000-hour rule”): practice with specific structural requirements that routine repetition usually lacks: (1) a well-defined, challenging goal just beyond current ability, (2) full, focused attention on the task rather than autopilot, (3) immediate, specific feedback on performance, and (4) repetition with refinement — repeatedly attacking the same weakness until it improves. Deliberate practice is effortful and often not enjoyable in the moment (contrast with Lesson 14’s “desirable difficulty” — same underlying principle, applied to skill-building rather than memorization) — which is exactly why routine, comfortable repetition is so much more common than genuine deliberate practice, even among people who practice a great deal.
- Transfer: whether skill or improvement gained in one task/domain carries over to a different task/domain. A robust and somewhat humbling finding across this research: transfer is much more limited than intuition suggests. Extensive chess expertise doesn’t reliably make someone better at general memory tasks or general strategic reasoning outside chess (consistent with Lesson 19: the advantage is chunking specific to meaningful chess patterns, not a general cognitive upgrade). Training “working memory” with generic memory-span exercises tends to improve performance on tasks very similar to the training task, but shows little to no reliable transfer to genuinely different cognitive tasks, despite popular “brain training” claims to the contrary.
The puzzle (MCQ above)
Think about what’s structurally different between “read 2,000 scans this year” (volume) and “read 2,000 scans this year, each time predicting the diagnosis, then checking the confirmed outcome and specifically studying the misses” (volume + deliberate practice’s key ingredients).
Part 2 — Why does routine practice plateau specifically, rather than continuing to slowly improve forever?
Lesson 18’s power law technically predicts some continued improvement at every additional trial, just a shrinking amount — never a hard plateau. Real-world routine practice, though, often produces genuine flat performance for years. Propose a reason routine, comfortable-difficulty repetition would produce an actual plateau rather than just an ever-slower-but-still-positive improvement curve. (Hint: think about what’s required for a trial to count as “practice that changes the underlying representation” versus a trial that’s just successfully executing an already-mastered chunk.)
Part 3 — Reveal: design a deliberate-practice regimen
Pick a skill you (or a hypothetical learner) might want to improve at — could be a professional skill (code review, technical writing, public speaking) or a hobby. Sketch a deliberate-practice regimen for it that concretely satisfies all four requirements listed above (challenging goal at the edge of ability, focused attention, immediate feedback, repetition with refinement) — and explain specifically what routine, non-deliberate practice at the same skill would look like by contrast.
Part 4 — Connect to modern agents
In a modern LLM agent harness, is there a meaningful analogue to “deliberate practice vs. routine repetition” — some way an agent (or the system training/fine-tuning it) could accumulate experience that’s structured like deliberate practice (targeting specific known weaknesses, with clear corrective feedback) versus experience that’s more like routine repetition (just running lots of tasks without targeted feedback)? What would you expect the “plateau” failure mode to look like for an agent stuck in the routine-repetition regime?