Nick Epley

Exposure therapy requiring individuals to engage in real-world social interactions is more effective at overcoming social anxiety than simulated or imagined interactions.

"If you are afraid of talking with a stranger or having a deep conversation, the way to get over that is not to simulate it or to imagine. It's not like you get up and you give a pretend speech. That's what psychologists were doing for years. It doesn't work because it's still pretending. It has to be real. You send people out in the world and to do the thing for real." (said at 0:00:00)

While real-world (in vivo) exposure is traditionally regarded as the gold standard and meta-analyses show a modest advantage in efficacy and long-term durability specifically for social anxiety disorder, simulated modalities (such as virtual reality exposure therapy) achieve comparable, large effect sizes immediately post-treatment.

• supports: Inferiority or Even Superiority of Virtual Reality Exposure Therapy in Phobias?-A Systemat… (Frontiers in psychology 2019)

  "Specifically, effect sizes for Specific Phobia (g = -0.15) and Agoraphobia (g = -0.01) were non-significant, only for Social Phobia we found a significant effect size favoring in vivo (g = - 0.50 )." (abstract, results)

  pubmedfull study (doi)
• context: Virtual reality exposure therapy for social anxiety disorder: a systematic review and meta… (Psychological medicine 2020)

  "When compared to in vivo exposure, the efficacy of VRET was similar at post-intervention but became inferior at later follow-up points." (abstract, results, passage verified)

  pubmedfull study (doi)
• context: A Meta-Analysis of the Efficacy of Virtual Reality and In Vivo Exposure Therapy as Psychol… (Behavior modification 2022)

  "Meta-analysis yielded a large significant effect wherein VRET resulted in significant reductions in PSA versus control of -1.39 ( Z = 3.96, p < .001) and a similar large significant effect wherein IVET resulted in significant reductions in PSA versus control of -1.41 ( Z = 7.51, p < .001). Although IVET was marginally superior to VRET, both interventions proved efficacious." (abstract, results)

  pubmedfull study (doi)

In economic experiments where participants are given money to allocate between themselves and an anonymous stranger, individuals typically share between 30% and 50% rather than keeping all of it.

"Like if I give you $10 in an experiment and tell you you can divide it with another person however you want—you can keep all of it, give none of it, or give it all away—the standard prediction from economic theory is you'll be purely self-interested. You won't care anything about this person you know nothing about. You'll give them nothing. That's not what real people do. They give something: 30%, 50% typically, depending on the context that you're in." (said at 0:49:40)

In behavioral economics experiments (notably the Dictator Game and Ultimatum Game), participants given money to split with an anonymous stranger reliably give a substantial portion of their endowment rather than keeping all of it. In classic Dictator Games (where the recipient cannot reject the offer), meta-analytic evidence shows participants transfer an average of approximately 20% to 30% of their endowment (with notable clustering at 0% and 50%/equal splits). In Ultimatum Games (where recipients can reject unfair offers), mean offers typically range from 40% to 50%. The claim correctly captures the robust finding that humans consistently display other-regarding preferences rather than pure self-interest, though the 30% to 50% range slightly overestimates the mean baseline giving in pure Dictator experiments.

• context: Gender differences in dictator giving: A high-power laboratory test. (PloS one 2025)

  "We gather information from a large laboratory sample comprising 1,161 subjects and study gender differences in altruism using a dual-role dictator game... females transfer 7.5 percentage points (about 40%) more of their endowment than males, on average." (abstract, results)

  pubmedfull study (doi)
• context: A publicly available benchmark for assessing large language models' ability to predict how… (Scientific reports 2025)

  "This benchmark consists of 106 textual instructions from dictator games experiments conducted with human participants from 12 countries, alongside with a compendium of actual human behavior in each experiment... identifying three major classes of behavior: self-interested, inequity-averse, and fully altruistic." (abstract, results, passage verified)

  pubmedfull study (doi)