Benjamin Perry @Benjamin_EPerry

I think this is one of the best examples illustrating how LLM Al is a purely statistical model and is not doing any kind of actual "thinking" or "logic" and cannot be trusted for tasks that require even a modicum of reasoning, a short 🧵:

Find me on bsky @colin-fraser.net @colin_fraser

2 years ago

A question indicative of (very approximately) 130 IQ

210 347 12K 5.5M 1K

It's new, it's improved, it's got AI-generated explanations for why everything correlates. Today I'm launching the new Spurious Correlations: tylervigen.com/spurious-corre…

What does it say about your IQ (g) if you prefer numerical (e.g., '20% chance of rain') versus descriptive (e.g., 'small chance of rain') weather forecasts? Attitude Toward Numbers (ATN), Need for Cognition (NFC) and Financial Literacy (FL): new paper tinyurl.com/4wksyrue

We Empirically Tested Astrology, And Here's What We Found Many people think that astrology is total bullshit. And yet many others swear by astrology, using it to understand themselves and the world and even buy related products. A YouGov poll found that 32% of Americans at least "somewhat agree" that someone's "astrological sign (e.g., Gemini, Pisces) accurately describes their character and personality traits." While many forms of astrology exist, the simplest formulation (sun sign / zodiac sign) astrology is quite easy to test empirically. So, we decided to put it to the test. We asked 323 people in the U.S. what their zodiac sign is and asked about 37 "life outcomes," including things like political alignment, religiosity, time spent volunteering, suicidal ideation, satisfaction with various areas of their life, BMI, and so on. Then, we trained a linear regression model (Ridge Regression, to be precise) to try to predict each of these 37 life outcomes (one at a time) using each person's zodiac sign (we represent each person by 12 numbers, with a 1 corresponding to their zodiac sign, and a 0 for all the other numbers since people each only have one zodiac sign). As a comparison point, we also measured people's personalities (using a "Big 5" personality test) and used those 5 personality trait scores (extroversion, conscientiousness, etc.) to try to predict these outcomes as well. How did astrology fair when we applied these models to predict life outcomes for new people on whom the model had not been trained? Absolutely horribly - it had no ability to predict the 37 outcomes. Compare this to personality traits, which predicted the outcomes with a reasonable level of accuracy (see the first image below). Varying the parameter of the prediction algorithm (known as "alpha," which tunes to what extent the model avoids overreacting to noise) didn't help either. Astrological sun signs were useless for predicting all 37 outcomes in the sense that making predictions for new people (that the model had not been trained on) was no better using astrological signs than simply predicting the average value for each life outcome (see the second image). So you'd be no more accurate using zodiac signs to predict these life outcomes than simply predicting that every person is exactly average on every outcome. Our results mirror those of OkCupid, which applied its dating compatibility algorithm to every pair of zodiac signs and found no meaningful difference in match scores across any of the pairs (see the third image). Of course, this work we did is only a test of sun sign astrology; it doesn't rule out the possibility that some of the more complex astrological systems work. But some of those systems still have sun signs at their core, so finding no relationships at all between sun signs and life outcomes is evidence against some (but not all) of those other systems working as well. You can read more details on the Clearer Thinking website (the link is in a comment below).

keyboard_arrow_left Previous keyboard_arrow_right Next

Can machine learning uncover the multivariate neural signature of major depressive disorder in individual patients? This large-scale study including 1,801 patients and controls finds no robust multivariate depression markers. ja.ma/3S0QgFY

Nearly three-quarters of tested psychological scales failed the test of measurement invariance needed for meaningful group comparisons psycnet.apa.org/record/2024-39…

1/22 Our new paper led by @ashleylwatts (w @ashlgreene & @wesbonifay) is now published; I view it as the first critical evaluation of the statistical and theoretical p-factor & resulting literature. Here a brief overview of the core arguments in the paper.

1/3 Most studies of the relationship between big 5 personality traits and political ideology examine the effects of broad or domain-level traits (e.g. 'Conscientiousness', 'Agreeableness'). However, the results of such studies can be misleading insofar as groups may have different or opposite scores on facets within those domain-level traits. For example, if liberals score higher than conservatives on the 'compassion' facet of agreeableness, and conservatives higher than liberals on the 'politeness' facet, these differences are likely to offset, resulting in a domain-level correlation with ideology that is small or indistinguishable from zero. Given this issue, a relatively recent paper opted to examine the domain-level AND facet-level relationships with ideology. The chart below (you'll probably have to zoom in) visualizes/summarizes the estimates of these relationships.

Rates of replicability across subdisciplines in psychology. Personality wins! 🧵pnas.org/doi/abs/10.107…

Building the Structure of Personality from the Bottom-Up using Hierarchical Exploratory Graph Analysis osf.io/zx7tc/

Best paper title 2023 so far: "Welcome to the Jangle: Comparing the Empirical Profiles of the 'Dark' Factor and Antagonism" journals.sagepub.com/doi/10.1177/10…

New Study🔔 Within-person and between-person structures of the Big Five personality trait measures. A recently published study has found that the within-person structure may be comprised of 2 factors. Read more here doi.org/10.1177/089020…

“Standardized tests are the best way to distinguish smart poor kids from stupid rich kids.” quillette.com/2023/12/01/the…

How do your favorite disorders fare in a data-driven reorganization of the DSM-5? Find out in this new preprint where we followed the patterns in 14,762 individuals’ symptoms with the goal of overcoming limitations in both the DSM and @HiTOP_system x.com/PsyArXiv_bot_v…

PsyArXiv bot v2 @PsyArXiv_bot_v2

3 years ago

Reconstructing Psychopathology: A data-driven reorganization of the symptoms in DSM-5 osf.io/7um9a/

0 18 50 105K 33

Spousal correlations for mental disorders are largely positive, meaning mentally ill people get with other mentally ill people more than expected by chance. This is important because people with two (vs 1/0) mentally ill parents are much more likely to develop mental disorders.

maximus @nuance_enjoyer

3 years ago

Psychiatric conditions and between-mate correlations. Autism, schizophrenia, and ADHD had the highest rates of assortative mating. Of course the skewed sex ratio makes this an unrealistic prospect for most autistic men though. sciencedirect.com/science/articl…

2 8 73 62K 30

validated doesn't mean valid validated doesn't mean good validated doesn't mean useful validated doesn't mean approved validated doesn't mean sufficient validated doesn't mean accurate validated doesn't mean best available validated means we did a test

Sex differences in basic vocational interests <0: Men more interested on average >0: Women more interested on average U.S. data; N = 1,283,110

1/2 In a new #OpenAccess study in BMJ Psych Open, we investigated the stability individual depression symptoms over a 9-year period in the NESDA data, with a focus on suicidal ideation. Fantastic effort led by Liia Kivelä here in @UniLeidenNews. cambridge.org/core/journals/…