Interview
#39 - Spencer Greenberg on the scientific approach to solving difficult everyday questions
- Plans to release the Uplift closed beta product shortly, followed by a public announcement and a new overconfidence prediction tool on clearthinking.org within a few months, with a calibration training tool scheduled for future release.
- Intends to utilize the Positively recruitment system using Mechanical Turk as the initial backend to accelerate social science research by lowering participant recruitment costs and complexity, while developing additional backends for varied sample types.
- Scheduled to analyze results from a long-term "happiness habits" study testing if triggers increase happy thoughts, with the potential to launch a public tool if outcomes are positive, and to investigate power posing effects on mood and feelings of power in an N=1,000 pre-registered study.
- Anticipates that power posing yields small, subtle effects consistent with six pre-registered replications (two positive, four null) rather than the original cortisol and risk-taking claims, though individual variation in posture-mood sensitivity may exist.
- Expectations regarding product and research risks include skepticism toward Theranos-style companies, categorizing eight out of ten as fraudulent due to delayed releases and lack of papers, and a general caution against specific dietary advice despite large individual base factors for rapid symptom relief.
- Forecasts on prediction markets and global security suggest election markets move gradually with polling as the primary prior, nuclear weapons have reduced great power war probabilities, and North Korea will likely retain weapons for strategic security rather than disarm.
- Cognitive biases and reasoning principles identified include the planning fallacy stemming from smoothed risks, underconfidence in difficult skills versus overconfidence in opinion-based traits, and the inability to reach 100% certainty through Bayesian updating.
- Methodological insights indicate reference class forecasting outperforms intuition by mitigating recency bias, while common sense fails in domains lacking immediate feedback, and the placebo effect is strongest on subjective measures where regression to the mean often explains observed improvements.
- Personal strategies mentioned involve listening to podcasts at speeds ranging from 1.2x to 2.3x depending on content density, and a warning to discount sensor data when it drastically contradicts personal sensation, implying equipment failure.
- Specific scientific myths and updates noted include the refutation of the "high iron" spinach decimal error legend, the realization that most of the "placebo effect" is actually natural recovery, and the observation that human brains are not rational machines given the difficulty of accurate reasoning.