Episode Summary
Executive Summary: Rob Wiblin interviews Spencer Greenberg about improving judgment: distinguishing intrinsic from instrumental values, using surveys to map what people value, identifying when people are over- or underconfident, and applying Bayesian thinking, reference classes, models, and evidence to real-world questions. The discussion also covers placebo effects, power posing, dieting claims, and how to stay rational without becoming alienated.
Main Topics: Spencer Greenberg’s projects and product portfolio: Greenberg updates on SparkWave’s portfolio companies: Uplift for depression, MindEase for anxiety, ClearThinking tools, and Positly for recruiting study participants. Intrinsic vs. instrumental values: The conversation centers on defining intrinsic values, surveying what people report valuing for its own sake, and why separating terminal from instrumental goals helps planning and self-understanding. Why identifying intrinsic values matters: Greenberg gives five reasons: avoiding value traps, better planning, reducing guilt/doublethink, and designing a compelling long-term vision for society. Overconfidence and underconfidence across skills: A study explores which skills people overrate or underrate, with five predictive traits such as perceived difficulty, experience, and whether a skill feels tied to personality. Bayesian updating and evidence: The episode explains Bayes factors, priors, odds-based updating, and the importance of not ignoring weak evidence or double-counting correlated evidence. Reference classes, models, and heuristics: The hosts discuss outside-view forecasting, planning fallacy, when to trust intuition, when to rely on models, and how qualitative heuristics can aid decisions like hiring. Skepticism, placebo, power posing, and disputed claims: They examine how to evaluate science and extraordinary claims, including power posing, placebo effects, dietary advice, Theranos, and why some surprising findings may still be partially true.
Key Arguments: Most people have multiple intrinsic values, not just one; effective altruists stand out mainly by emphasizing the suffering and happiness of all conscious beings. People often confuse values that matter instrumentally with things they value for their own sake, which can lead to career traps and inefficient life choices. Understanding your intrinsic values can reduce guilt and self-deception by clarifying that your current motivations may differ from your moral ideals. Many overconfidence findings are domain-specific: people are overconfident in some skills and underconfident in others depending on traits like ambiguity, experience, and perceived difficulty. Evidence should be assessed by the likelihood ratio: how much more likely is the observation if the hypothesis is true than if it is false? Bayesian reasoning implies beliefs should update gradually, and no belief should be treated as 100% certain unless it was already certain from the start. Reference classes are useful for priors, but they must be chosen carefully to balance relevance against sample size and avoid biased forecasts. Models are always wrong in some cases, so the best approach is often to compare multiple models and average their insights rather than rely on one theory. Scientific studies are only one type of evidence; intuition, direct observation, heuristics, and expert judgment all contribute, but each has different reliability depending on feedback quality and domain. Many surprising claims fail because of publication bias, poor controls, or overinterpretation, but some apparently disputed effects may still exist as small or variable effects.
Data Points: Uplift users: Used by people every week in closed beta - Spencer says the depression app is already in use and nearing public release SparkWave portfolio companies: 4 - Uplift, MindEase, ClearThinking, and Positly ClearThinking users: 150,000+ - Free tools and training programs have been used by over 150,000 people Happiness habit effect: 5% increase (self-reported) - Greenberg says a gratitude-trigger habit increased his happiness over several months Daily happy thoughts target: 20-30 extra thoughts - Goal of the happiness-habit intervention Non-EA sample size for intrinsic values study: 82% - Reported “I love other people” as an intrinsic value Non-EA sample size for intrinsic values study: 74% - Reported “I myself feel happy” as an intrinsic value Non-EA sample size for intrinsic values study: 71% - Reported “beautiful things continue to exist” as an intrinsic value Non-EA sample size for intrinsic values study: 69% - Reported “people I know personally feel happy” as an intrinsic value EA sample trait: 84% - Reported “people I know personally feeling happy” as intrinsic value EA sample trait: 81% - Reported “animals feel happy” as intrinsic value EA sample trait: 80% - Reported “I suffer a lot less than I do normally” as intrinsic value Question count in overconfidence study: 100 skills - Participants rated themselves on 100 skills relative to peers Knitting a sweater: Underconfident - Example skill where participants rated themselves below average Thinking critically: Overconfident - Example skill where participants rated themselves above average Lifting 10 pounds: Overconfident - Another skill where participants rated themselves above average Running a marathon: Underconfident - Example skill where participants rated themselves below average Overconfidence traits identified: 5 traits - Predictors included average perceived skill, subjectivity, experience, personality relevance, and difficulty EA survey filter: More than one quiz question wrong excluded - Participants who failed comprehension checks were removed Reference factor example: 3:1 - Likely evidence strength discussed for Bayes factors Strong evidence example: 30:1 - Used to illustrate strong Bayes factors World Cup example prior: 33% - Used in Bayes rule example before a goal update World Cup example posterior: 60% - After a goal, probability of winning rose from 33% to 60% in the example U.S.-China war historical reference class: 9 out of 13 - Example of great-power transitions going to war historically Trump-Russia update example: 10% to 30-40% - Illustrative posterior after considering later evidence Theranos-style fraud prior: A few percent - Greenberg’s rough prior that a new biotech breakthrough is outright fraud Prediction market/election example: Dozens of presidential elections - Used as a limited reference class for election forecasting Power posing replication result: 6 preregistered replications; 2 positive, 4 negative - Discussed as evidence for a small but possibly real effect Calibration quiz: 30 items - Common-misconception tool on ClearThinking Hiring resume correlation: Basically zero - Two evaluators’ resume ratings were uncorrelated Mechanical Turk backend: First backend - Initial recruiting platform for Positly Bayesian evidence phrasing: Likelihood ratio - How much more likely the evidence is if the hypothesis is true than if false
Pivotal Quotes: "An intrinsic value is something you value for its own sake, not for the effects that it has." — Spencer Greenberg: Definition introduced during the discussion of values "The way evidence should work is that we're kind of smooth. It's like gradually adjusting all the time." — Spencer Greenberg: Explaining Bayesian updating and why small evidence should not be ignored "A bad argument doesn't make something false. That's the fallacy fallacy." — Spencer Greenberg: Advice on handling alienating or irrational-sounding conversations
Implications: Listeners should expect better decisions if they clarify their true values, use outside views and Bayes-style updating, and remain open to small but cumulative evidence. The episode argues for epistemic humility, plural evidence, and practical rationality without perfectionism.