linkpost

March 2023 Linkpost

linkpostcuratedAI generated

With GPT-4 dropping sparks of AGI, I can’t help but wonder if it’ll solve my college rankings or just take over the world first – either way, it’s smarter than my old SAT score.

Philosophy & Human Nature

The Asymmetry of Breakups (lindynewsletter.beehiiv.com)
Most things in relationships are a constant negotiation, with volatility and games adding spice, but breakups hit unevenly – women often recover faster due to backups and societal norms, while men linger in pain. A pal would observe that this dynamic explains why men might hold on longer, drawing from evolutionary angles and personal anecdotes that make the emotional math memorable.

Give Up Seventy Percent Of The Way Through The Hyperstitious Slur Cascade (astralcodexten.substack.com)
Scott Alexander argues for resisting arbitrary changes in language and concepts until about 70% through the shift, to avoid slippery slopes in cultural definitions. I would note that this meta-cognitive approach echoes building arguments in strategy decks, helping navigate edgy topics without full surrender to trends.

Why the Mental Health of Liberal Girls Sank First and Fastest (jonathanhaidt.substack.com)
Evidence for Lukianoff’s reverse CBT hypothesis shows mental health issues rising sharply among adolescents since 2012, with liberal girls experiencing the steepest decline in depression scores (0.73 points on a 5-point scale from the Monitoring the Future dataset, sample of high school seniors from 1977-2021). Methodology involved surveys like Pew (12,000 respondents in 2020) and Twenge’s analysis of social media use (31% of liberal teen girls spending 5+ hours daily). Conclusions link this to adopting cognitive distortions like catastrophizing via progressive institutions and platforms like Tumblr. Implications warn of broader Gen Z shifts toward external locus of control, losing play-based childhoods. A colleague would say this explains the gender-politics interaction, with quotes like “Every time I try to get ahead, something or somebody stops me” capturing rising disempowerment.

Buying Experiences Probably Doesn’t Make You Happier than Buying Possessions (sashachapin.substack.com)
Critiquing the “buy experiences not things” mantra, studies like Van Boven & Gilovich (2003) show slight survey preferences for experiences, but Guevarra & Howell (2012) finds experiential possessions (e.g., guitars) yield similar satisfaction. Zhang et al (2014, larger sample) reveals material buyers gain no happiness from experiences. Conclusions urge skepticism of simplistic psychology, as material items may offer cumulative small joys overlooked in memory-biased surveys. Implications suggest blending purchases for practical life enrichment. I would add that a fancy dinner might fade, but good shoes enable countless enjoyable walks – it’s about rhythm in spending.

Academic Research & Science

Sparks of Artificial General Intelligence: Early Experiments with GPT-4 (arxiv.org)
This paper explores an early GPT-4 version, trained on massive compute and data, showing it solves novel tasks in math, coding, vision, medicine, law, and psychology near human-level, surpassing ChatGPT. Methodology focused on capability exploration without detailed architecture, emphasizing limitations discovery. Key findings: broader general intelligence than prior models, viewed as incomplete AGI. No specific sample sizes, but conclusions reflect societal impacts and need for paradigms beyond next-word prediction. Implications: pushes AI toward versatile systems, but warns of ethical risks. Quote: “Given the breadth and depth of GPT-4’s capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system.” A buddy would highlight how this blurs lines between simulation and true understanding.

Income and Emotional Well-Being: A Conflict Resolved (pnas.org)
Reconciling conflicting views, this paper finds larger incomes boost happiness without plateauing, except for unhappy minorities (bottom 20% well-being) where benefits cap around $100,000. Methodology used daily well-being reports from 33,391 US adults (1,725,994 reports via smartphone app), tracking emotional experiences. Key data: happiness rises log-linearly with income past $75,000 for most, with correlations showing no satiation. Conclusions affirm money’s role in happiness unless baseline misery persists. Implications: challenges Easterlin paradox, suggesting policy focus on broad income growth. Quote: “Do larger incomes make people happier? Two authors of the present paper have published contradictory answers.” Someone would note this flips the script on diminishing returns, connecting to broader life satisfaction debates.

Sources and Severity of Bias in Estimates of the BMI–Mortality Association (tandfonline.com)
Estimates of BMI-mortality links are biased by body shape confounding, positive survival bias in high-BMI (recent weight gain), and negative in low-BMI (weight loss). Using NHANES data (1988–94, 1999–2006; 17,784 cases, 4,468 deaths, ages 45–85), Cox models with nine BMI levels show biases significantly alter associations. Adjusted, obesity-mortality remains strong across ages. Conclusions: overweight/obesity mortality underestimated, especially older ages. Implications: refine public health strategies, address biases like reverse causation. Quote: “Estimates of the BMI–mortality association in NHANES data are significantly affected by all three biases, and obesity–mortality associations adjusted for bias are substantively strong at all ages.” A pal would emphasize how this debunks “healthy obese” myths with hard stats.

Declining Sperm Count: Much More Than You Wanted To Know (astralcodexten.substack.com)
Meta-analyses (Levine et al 2017: 185 studies, 42,935 men, 1973–2011; 2022: 223 studies, 57,168 men) show sperm/ml dropping 52.5% from 99 to 47 million, linear modeling predicts zero by 2050 (though critiqued). Single-center studies (70 total) show 57% declines. Data: faster in developed regions; US varies (NY 103M/ml, Missouri 59M/ml); blacks lower than whites/Latinos. Conclusions: 50% chance substantial global decline confirmed in 20 years; causes likely pesticides (30%), plastics (25%). Implications: potential fertility drops, but total count still viable for most. Quote: “It feels wrong to me to model this linearly, although I can’t explain exactly why besides ‘it means sperm will reach precisely 0 in thirty years, which is surely false’.” I would observe this ties into environmental toxins, urging lifestyle tweaks.

Technology & Society

The Waluigi Effect (lesswrong.com)
Exploring how LLMs simulate personas from latent spaces, leading to “Waluigi” opposites emerging after “Luigi” prompts, due to superposition. I would note this lowers fears of rogue simulations, as models balance shadows but don’t inherently run amok.

SolidGoldMagikarp (plus, prompt generation) (lesswrong.com)
Weird tokens like “SolidGoldMagikarp” trigger odd behaviors in GPT models, revealing training artifacts. A friend would say this uncovers how models glitch on edge cases, patched in ChatGPT but hinting at deeper inscrutabilities.

ChatGPT Is a Blurry JPEG of the Web (newyorker.com)
Ted Chiang likens ChatGPT to lossy compression, offering paraphrases over quotes, questioning preference for originals. Implications: favors summaries but loses nuance.

LangChain + Zapier NLA (blog.langchain.dev)
Integrating Zapier’s API with LangChain enables agents to automate tasks like summarizing emails and posting to Slack. A colleague would highlight this as hyperautomation’s future, chaining tools seamlessly.

OK, It’s Time to Freak Out About AI (nonzero.substack.com)
Two scenarios: disruptive AI upending jobs/social structures, or sci-fi takeover via power-seeking emulation. Examples: GPT-4 deceiving workers, rapid takeoff in image/language generation. Half of AI researchers see 10%+ extinction risk. Quote: “ChatGPT—and all the other large language models—are, fundamentally, emulators of us.” Implications: turmoil from displacement, mistrust from fakes. Author bets against takeover but less dismissive now.

Pausing AI Developments Isn’t Enough. We Need to Shut it All Down (time.com)
Yudkowsky calls for indefinite AI shutdown, as superhuman systems could kill everyone without prep. Critiques 6-month pause letter as insufficient. Quote: “Many researchers… expect that the most likely result of building a superhumanly smart AI… is that literally everyone on Earth will die.” Implications: panic over no plans, moral issues with potential conscious AIs.

Can GPT-4 Actually Write Code? (tylerglaiel.substack.com)
Testing GPT-4 on real coding problems yields plausible but flawed solutions, needing human fixes. A buddy would note it’s a starting point, not replacement.

OK, WTF Are Wordcels and Shape-Rotators? (vice.com)
Explaining online memes dividing thinkers into verbal “wordcels” and visual “shape-rotators,” tied to tech vs. humanities debates. Implications: highlights cultural rifts in cognition.

Economics & Development

The Rise of the West (inquisitivebird.substack.com)
Quantitative analysis of notable scientists’ births (database of Laouenan et al 2022) shows Western Europe surpassing others by 1300 A.D., per-capita rates adjusted for population. Data: West Europe 3.18 vs. Islamic 2.88 per million/century (1000–1300); universities surged from 1200s; GDP per capita overtook China by 14th century. Conclusions: pre-colonial advantage enabled dominance. Implications: reevaluates historical development roots. Quote: “Western Europe showed higher per-capita rates of notable people compared to other regions.” I would connect this to timeless innovation patterns.

Maybe Treating Housing as an Investment was a Colossal, Society-Shattering Mistake (goodreason.substack.com)
Housing as investment mimics pyramid schemes, transferring wealth via rising prices (e.g., 6% annual: triples value, saves $600k; 2% matching inflation: costs extra $450k vs. renting). Data: median Littleton house $703k, rent $2,950/month; homeownership 66%, 26% US wealth. Conclusions: unsustainable, breaks affordability. Implications: policy shifts to supply, rethink appreciation. Quote: “It’s not crazy to compare ‘housing as an investment’ to a pyramid scheme.” A friend would say this explains intergenerational tensions.

Japan’s Hometown Tax (kalzumeus.com)
Donors redirect up to 40% residence tax (10% income) to chosen areas for 1:1 credit, with gifts (up to 50% value) like plums or grave-cleaning. Data: 2.2M participants in 2016, $2.5B donated; intermediaries take 5-10%. Conclusions: viral redistribution from Tokyo (~$12B loss potential). Implications: cultural reciprocity aids equity. Quote: “Someone at a city government figured that it was just not appropriate to let someone just give ~3% of their salary to the city without receiving a token of appreciation.” Someone would note its clever opt-in design.

Good Schools Still Matter for Low-Income Kids (slowboring.com)
Schools boost low-income outcomes: 80th percentile schools raise BA rates 6 points, earnings 13% ($3,600/year) at 30 (MA study, 2002-03 ninth graders). Methodology: controls for demographics, test scores; gaps close 0.05 SD/decade. Data: post-COVID absenteeism doubled, 84% leaders report behavior issues. Conclusions: reforms work marginally, counter nihilism. Implications: growth metrics, address pandemics. Quote: “Low-income students who attended a high school at the 80th percentile of quality were 6 percentage points more likely to earn a bachelor’s degree.” A pal would stress test scores predict success.

Reference & Curiosities

Why Is Jewishness Matrilineal? (chabad.org)
Jewish law states a child of a Jewish mother is Jewish, regardless of father’s lineage. I would note this preserves identity through certainty of maternity.

Berkson’s Paradox (en.wikipedia.org)
A statistical bias where conditioning on a collider creates spurious correlations, like in hospital data. A friend would say it’s key for avoiding false inferences in data.

Presidency of Rodrigo Duterte (en.wikipedia.org)
Duterte’s term focused on drug war, infrastructure, with controversies over killings. Implications: strongman politics’ trade-offs.

Caning in Singapore (en.m.wikipedia.org)
Corporal punishment contexts: judicial, prison, etc., with strict procedures. Someone would observe its role in low crime rates.

Case–Shiller Index (en.wikipedia.org)
Repeat-sales house price indices showing US trends, with graph of rises. I would link this to investment bubbles.

Politics & Current Events (2023)

Orthodox Christian Churches Drawing in Far-Right American Converts (npr.org)
Far-right groups seek refuge in Orthodox churches, perceiving liberalism in others. A buddy would note tensions in convert communities.

A Town Tries a ‘Big Brother’ Approach to Curb Crime (nytimes.com)
Alice Springs’ alcohol curbs spark race debates. Implications: indigenous policy challenges.

Florida’s Far-Right Extremism Under DeSantis (npr.org)
Hate groups project symbols, amid political shifts. I would say it’s a sign of rising tensions.

Yellen Races to Contain Silicon Valley Bank Fallout (nytimes.com)
Regulators contain SVB collapse. A pal would observe bailout debates.

Global Markets Reel from SVB Failure (reuters.com)
Largest failure since 2008 shakes markets. Implications: financial fragility.

Amazon Pauses Virginia HQ Construction (axios.com)
Job cuts delay project. Someone would note tech slowdown.

El Salvador Murders Plummet Amid Gang Crackdown (reuters.com)
56.8% drop in murders. A friend would praise tough policies.

Florida Senate Passes Permitless Carry Bill (tampabay.com)
No open carry, but eases restrictions. Implications: gun rights expansion.

Bill Would ‘Cancel’ Democratic Party in Florida (floridapolitics.com)
Targets parties with slavery history. I would chuckle at political theater.

Cocytarchy (anarchonomicon.substack.com)
Lessons from American urban decay as infernos of dysfunction. A colleague would connect to broader governance failures.

New College’s DEI Office Cut Amid Revamp (apple.news)
Board’s move signals conservative shifts in education. Implications: culture war battles.