TB: Lifelong Learning

Overview

Learning doesn't stop at graduation — it compounds. The people who stay intellectually alive into old age are those who treat curiosity as a practice and who arrange their lives to encounter novelty constantly. At the same time, formal education is largely a signaling game: its primary output is a credential, not competence. Understanding this distinction clarifies what is worth pursuing and what isn't.

Core Tension

Formal education is expensive, time-consuming, and mostly signaling — yet credentials still function as gatekeepers to important opportunities. The resolution is to treat formal credentials as necessary minimums (where required) while building real competence through deliberate self-directed learning. Don't mistake the certificate for the education.

Key Insights

Education is mostly signaling

  • "Sheepskin effect": most of the earnings premium from education is captured at the moment of graduation, not accumulated over the years of study. A Princeton diploma is more valuable than a Princeton education without the diploma — which suggests the value is the signal, not the knowledge.
  • Schools teach subjects that aren't useful to employers and omit many that are. The curriculum is not designed around competence; it's designed around measurable, defensible criteria.
  • What education signals: intelligence (you can pass hard tests), work ethic (you completed the program), and conformity (you played by the rules for four years). Employers care about all three.
  • Education's position in society is robust to technological disruption because unconventionally signaling conformity (e.g., "I'm smart enough not to need a degree") paradoxically signals nonconformity. First movers who skip credentials face steep costs, and collective action to change the system is very hard.

Teaching is one of the best ways to learn

  • Explaining something forces you to identify gaps in your own understanding. The Feynman technique (explain it simply; if you can't, you don't understand it) works because the act of explanation is a form of testing.
  • Teaching also forces organization: you have to find the conceptual skeleton before you can dress it in examples. The preparation is as educational as the delivery.

Children as a mode of education

  • Having and raising children provides a peculiar form of education: you encounter a mind that hasn't yet accepted the assumptions you take for granted. Children ask the embarrassing questions (why?) that reveal how shallow most adult understanding is.
  • The parental imperative to answer forces you to actually know things, not just recognize them. It is a perpetual Feynman test.

AI as the new learning frontier

  • Learning to be productive with AI is arguably the most important learnable skill right now. It's the skill of how to work with and direct a technically capable intelligent system.
  • AI provides: massive productivity leverage, a new mode of exploration (socratic dialogue with a vast knowledge base), and a new way to think (forcing you to articulate questions precisely).
  • The key AI-era skill is not knowing facts but knowing how to ask well, evaluate answers critically, and integrate AI output with your own judgment.

GAI in the undergraduate STEM classroom

  • Stubbs et al. (arXiv:2602.04389, Harvard Astro 17, Fall 2025): fully integrated GAI (Gemini suite) into a 14-student introductory astronomy course. Key finding: utility is highly use-case-dependent. Code generation and debugging (mean score 4.40/5) was the clear sweet spot; open-ended self-guided study (2.60/5) was least effective.
  • The single most transferable lesson: explain the rationale for GAI on day one. Students who lacked explicit framing were confused and disengaged all semester; a single explanation mid-semester transformed their engagement. Students do not share instructors' conviction that AI fluency is a critical emerging skill — this must be argued, not assumed.
  • Hybrid assignment design: each problem set had explicit AI-OK and no-AI questions. No-AI questions were flagged as previewing exam content — this incentivizes genuine mastery without any compliance monitoring. No-electronics proctored exams enforced baseline competence.
  • Public GAI channels (RAG Slack bot) failed: students strongly prefer private 1:1 interaction. Harvard Business School confirmed the same finding independently. Do not invest in shared AI channels when individual access is available.
  • Unguided self-study with GAI is least effective for early-stage learners: they lack the domain knowledge to navigate open-ended dialogue productively. The Feynman test applies — you must already understand enough to recognize a wrong answer.
  • Instructor-side wins: rapid custom figures from natural-language prompts; interactive Colab notebooks with real SDSS/Rubin data (previously infeasible); GAI "taking" draft exams to surface ambiguous wording before students see them.
  • Course evaluations did not decline: mean 3.44 (2024, no GAI) → 3.91 (2025, with GAI). ~54% agreed "AI aspects enhanced my learning." Students valued human feedback and instructor presence; GAI did not undermine human-centered aspects.

Related Concepts

Cross-Topic Connections

Contributing Articles

Backlinks