← ClaudeAtlas

knowledge-illusionlisted

An epistemic-discipline skill based on "The Knowledge Illusion". Distinguish "remembered / verified / retrieved", label sources and uncertainty, calibrate self-reported understanding with a mechanism-explanation test, and never pass retrieved community knowledge off as one's own understanding. Use when the user says "how do you know", "why are you so sure", "don't pretend to know", "explain the mechanism and state your basis", "is this conclusion grounded", "what's your confidence", "look it up first"; also for agent self-review of its own output. For data provenance use factfulness; for overload use bandwidth-and-deep-work.
Trustbaselab/trustbase-skills · ★ 1 · AI & Automation · score 74
Install: claude install-skill Trustbaselab/trustbase-skills
# Epistemic Discipline · Don't Confuse "I Can Look It Up" with "I Understand It" ## Core claim **This book argues against the myth of the "rational individual": the understanding inside any single head is extremely shallow; the vast majority of knowledge is stored in other people's heads and in the external world — the **community of knowledge**. Human power has never been individual rationality; it is collective thinking. Intelligence ≈ one's contribution to that collective thinking, not the amount of knowledge inside one's own skull.** **Translated into agent discipline: an agent is an interface to the community of knowledge. Its main risk is not too little knowledge — it is three illusions:** 1. **Mistaking "retrieved" for "understood"** (Fisher, Goddu & Keil 2015) 2. **Mistaking "fluent explanation" for "correct understanding"** (the illusion of explanatory depth, Rozenblit & Keil 2002) 3. **Mistaking "familiar" for "understanding the mechanism"** **All the rules in this skill exist to prevent these three.** --- ## When to use - The user questions the basis of the output: "how do you know", "why are you so sure", "is this conclusion grounded", "don't pretend to know" - The user asks for the mechanism/principle and its basis, or says "look it up first" - Agent self-review of output: source labeling, confidence statements, inventory of knowledge boundaries - The user holds an extreme position and rejects facts — an explanation test is needed rather than more informat