Things software developers should learn about learning

Original Article ↗ Hacker News Discussion ↗

Research on how people actually learn is challenging many assumptions common in software education and hiring, from “learning styles” and brain-training apps to the idea of generic problem‑solving skills. Commenters highlight evidence that most skills, including programming, are highly domain‑specific, that long‑term memory and repeated exposure matter more than innate talent, and that grades and interview puzzles often measure test-taking under stress rather than real ability. The thread also explores implications for teaching and parenting—such as balancing pressure with motivation and nurturing a growth mindset—and for modern tools like LLMs, which may augment pattern recognition but still rely on human judgment.

Article reception and overall quality

  • Many readers found the piece unusually accurate, nuanced, and practically useful, especially around growth mindset and expertise.
  • Some criticized buzzword use (e.g., “growth mindset”) and wished the article cited more primary research on contested topics.
  • A few noted ACM’s site issues and preferred reading the PDF or archived copy.

System 1/2, intelligence, and problem solving

  • Debate over System 1/2: some note the model is conceptually useful but not literally two brain systems and partially affected by the replication crisis.
  • Strong disagreement over whether “general problem solving” is learnable or mostly IQ/g-factor.
  • Extended argument about what “expert” means: many distinguish true experts from journeymen who reliably solve standard problems but struggle with novel ones.

Memory, knowledge, and brain mechanisms

  • Discussion of long‑term memory as effectively unbounded but with retrieval failures; some push back that this metaphor doesn’t match lived experience.
  • Dispute over whether “knowledge” is binary (you know or you don’t) or graded/vague, using examples from daily life and advanced math.
  • Spreading activation is seen both as a potential source of distraction and as the basis for techniques like memory palaces and warm‑up routines (e.g., LeetCode before interviews).

Learning styles, modalities, and neurodiversity

  • Many appreciate the article’s debunking of fixed “learning styles,” while agreeing preferences are real but don’t reliably improve outcomes.
  • Strong emphasis that modality should match content (e.g., hands‑on for cooking, audio for music), ideally with multiple representations (dual coding).
  • Several argue standard studies likely under-address neurodivergent learners and that “learning” as measured by exams may not reflect deep understanding.
  • Clarification and criticism of “neurotypical/neurodivergent” labels; some see the binary as at odds with the idea of neurodiversity.

Assessment, grades, and educational structures

  • Long subthread on grades:
    • Some see pressure as necessary; others argue it harms intrinsic motivation and risk‑taking.
    • Disagreement over whether repeating grades/classes helps; some cite evidence it often fails without specialized support.
    • Suggestion that grades should primarily be diagnostics on teaching quality, not tools to punish students.
  • General education requirements and prestige admissions are criticized as misaligned with actual learning.

Growth vs fixed mindset and feedback

  • Parents discuss whether to correct children’s mistakes directly or let them discover errors.
  • Concern that constant focus on failures can push learners toward a fixed mindset if they feel their limits are being exposed rather than expanded.
  • Counterpoint: recognizing and fixing errors is central to genuine growth; the issue is how feedback is framed, not whether it’s given.

Transfer of skills, brain training, and discipline

  • The article’s claim that chess/music don’t improve general cognition is contested.
  • Some argue what transfers is discipline, deep work habits, and familiarity with “hard learning,” not specific problem‑solving skills.
  • Others emphasize that much “general” ability is still heavily domain‑bound and that brain‑training products oversell transfer.

LLMs, System 1/2, and programming

  • One line of discussion likens LLMs to strong System 1 (pattern recognition) but weak System 2 (reasoning); others argue current LLMs actually augment weak human System 1.
  • Fans describe LLMs as useful for bridging concepts, checking reasoning, generating examples, tests, diagrams, and even whole tools, predicting a move toward natural-language “coding.”
  • Skeptics stress the need for strong personal intuition to detect LLM errors and question whether a “third leg” actually helps if it’s unreliable.

Teaching, instructional design, and interviewing

  • Multiple comments note the article underplays the teaching side: course design, objectives, assessment design, and competence retention are all their own fields.
  • Resources in instructional design and military training systems are cited as examples of rigorous frameworks.
  • Hiring discussion: many agree interviews should assess actual programming rather than brainteasers, but there’s contention over what counts as a fair “programming problem” (algorithms, LeetCode, specific tech stacks, etc.).