• “The Algorithm Will See You Now” () - Context: Radiology is medicine’s most digitised domain - perfectly positioned for AI - but progress toward human replacement has been oddly sluggish. ! - Performance paradox: CheXNet and its successors can outperform radiologists on lab benchmarks, yet models often collapse in real hospitals due to narrow training data, calibration gaps, and bias in under-represented demographics. - Regulatory drag: FDA approval demands separate validation for each retrain; malpractice insurers routinely write “absolute AI exclusions”. Hospitals thus keep humans “in the loop”. - ‍ Task granularity: Only a third of a radiologist’s job is image interpretation - the rest is counselling, protocol design, and coordination - so automation hits diminishing returns quickly. - Elastic demand: As imaging got faster and cheaper with digital systems, scan volumes rose 60%. AI could similarly amplify throughput rather than eliminate jobs. - Moral: Radiology’s lesson is sociotechnical, not computational - the better the machines, the busier the humans. - The Algorithm Will See You Now – Works in Progress
  2. Scott Alexander - “The Fatima Sun Miracle: Much More Than You Wanted To Know” (Astral Codex Ten)
    • Setup: In 1917, ~70,000 people at Fátima, Portugal, reported seeing the sun spin, flash colours, and plunge toward Earth - perhaps the most documented “miracle” in modern times.
    • Method: Alexander reconstructs 150+ testimonies from parish, diocesan, and newspaper archives - cross-checked via the Documentação Crítica de Fátima - to test whether the event could be natural, optical, or sociological.
    • Findings: Descriptions vary - spinning discs, coloured rays, descending lights - but no consistent meteorological or astronomical correlate emerges. Perception, expectation, and optical illusion intertwine.
    • Epistemic lens: Crowds can synchronise beliefs under uncertainty; “mass testimony” isn’t strong evidence when witnesses share context, cues, and anticipation.
    • Meta-lesson: Belief formation under ambiguity mirrors modern cognitive bias - collective sincerity is not collective reliability.
    • The Fatima Sun Miracle: Much More Than You Wanted To Know – Astral Codex Ten
  3. Economic Freedom & the Bottom 10%
    • Observation: Across nations, higher economic freedom correlates with higher absolute incomes for the poorest decile. A scatter plot graph titled "Economic Freedom and Bottom 10% Income Threshold." The x-axis shows economic freedom scores ranging from 0 to 10, divided into categories like "Least Free Average" and "Freest Average." The y-axis displays income levels in 2021 US dollars, adjusted for purchasing power parity, ranging from $0 to $18,000. Data points represent countries such as Norway, Switzerland, and Bangladesh, plotted with varying colors indicating economic freedom levels. Text labels include country names and income values for the bottom 10% of the population.
    • Scale: The bottom 10% earn roughly 7–8× more in the most-free economies than in the least-free - an absolute uplift even if their share of national income is unchanged.
    • Mechanism: Freer markets often coincide with stronger property rights, business formation, and trade openness - raising productivity that reaches the poor.
    • Caveat: Correlation isn’t causation - human capital, governance, and culture co-move - but institutional quality appears a reliable predictor of welfare gains at the base.
    • Policy signal: Protect the rule of law and market dynamism alongside redistribution if the goal is to raise real living standards for the poorest.