Wow this is an impressive article from @verawil.de that I highly recommend. One question-how can one get away with "artificially censored at the moment her actual behavior diverges from the assigned strategy" without making some kind of severe independent censoring assumption? #StatsSky #EpiSky

Yo dawg heard you like worrying about the implications of Bayes rule in mass screenings for low-prevalence problems. But have you tried worrying about intention-to-treat vs per-protocol estimates? Let's frolic thru the latest in target trial emulation! wildetruth.substack.com/p/its-the-an...

Hash-matching for known child sexual abuse material (CSAM), aka Chat Control 1.0: doable ✓ Mass AI scanning for unknown CSAM/ Chat Control 2.0: not ✗ Privacy advocates: 😭 mass AI scanning can work?! Child safety advocates: 😭 mass AI scanning can't work?! Me: yes, these are different problems

Updated my working paper "When Mass Security Screening Backfires" with a formal proof (§5.4) showing why hash-matching of known CSAM is resource-tractable, while mass AI scanning for unknown CSAM is mathematically intractable at scale, regardless of how good the classifier gets. tinyurl.com/twf538y3

A support group for AI-induced psychosis finds 3 delusions recur (nicknames my own): 1. The Frankenstein delusion - I've created life (consciousness) 2. The Napoleon Hill delusion - I'm going to get rich 3. The Ezekiel delusion — I've been chosen (Ezekiel 1) www.theguardian.com/lifeandstyle...

me at 3 a.m.

Oh hey Bayesian search is the right tool for the closed universe searches job, no? I mean if you know there's a crime-specific true positive among the false, and have secondary investigative ways of winnowing (tuning parameter optimization)

lol most of the replies to this are basically “you sweet naive child, you think 30 open tabs is a lot”

Imagine there’s no endogeneity It’s easy if you try No hidden selection bias No omitted reasons why Imagine all the scatterplots Meaning what they show

In a low stakes, artificial factorial experiment employers do not discriminate against immigrants. But when it is about actually hiring people, employers do prefer applicants with German-sounding names in the same experimental choice set.

I am told this Berlin Philharmonic version of Mahler's Symphonie No. 5 was performed entirely without sheet music, as is conductor Gustavo Dudamel's wont. www.youtube.com/watch?v=5J5x...

A new crypto paper shows how to evade mass AI scanning for child sexual abuse images (CSAM) during ongoing EU policy debates about different implementations of such scanning. Science should serve the public interest, and this doesn't. wildetruth.substack.com/p/when-secur...

Readings on the affect effect & description-experience gap; and redesigning my risk literacy app following feedback from brilliant risk literacy folks at Charité. German compound word power shines light in the darkness! wildetruth.substack.com/p/old-rarity verawilde.github.io/rarity-roule...

Oh yes German compound word creator oh yes necessary www.yeschat.ai/gpts-9t55R6R... I give you: Schimmerhoffnungsverzweiflung (an oddly hopeful despair) or (metaphorically): Schönheitstodschmetterling (beautiful death butterfly)

writers: can you believe some venues don’t even pay you for publishing with them academics:

Meanwhile, in the corner at a "just add Bayes!" party...

What's going on this week in mass screenings for low-prevalence problems? EU Chat Control trilogues, a US con law perspective on "closed universe searches," and a tentatively hopeful UK hepatitis C screening proposal. wildetruth.substack.com/p/mass-scree...

Oh wow, the EU Parliament voted to sunset Chat Control 1.0 today. That's the regime scanning some private services (e.g., Gmail) but not E2E (e.g., WhatsApp) for known child sexual abuse material ("CSAM"). Margin was narrow. Reg: www.patrick-breyer.de/wp-content/u...

New memoir working title: "Little Miss Can't Be Right: Imperfect estimates of one methodologist's biases and errors"

The EU votes tomorrow on mass AI scanning for known child abuse images. I support the sunsetting program. I'm also publishing a preprint suggesting its more expansive sister proposal, still being debated this summer, would catastrophically backfire. wildetruth.substack.com/p/the-eu-par...

New memoir working title: Base rate neglected: A mother's guide to high-impact, low-prevalence disasters

New memoir working title: "All Veras are wrong, but some are useful: A quantitative methodologist seeks meaning in science service"

New memoir working title: "All Veras are wrong, but some are useful: A quantitative methodologist seeks meaning in science service"

The nights remain cold, but the days are sunny and warm in Berlin at long last. Fast responders, the sprinkle blossoms are already blooming.

Surveillance folks: EU Parliament's LIBE committee votes Wednesday on extending Chat Control 1.0 — a voluntary AI mass scanning regime intended for child sexual abuse material (CSAM), set to expire April 3. What my working paper finds about why such screening backfires. 🧵

bro just one more future study bro, bro I swear just one more future study and it'll fix the inference bro

My parents got me this science chick necklace a few Christmases ago, and now that I finally wear it, man on the street says I have artist vibes. The math → art pipeline is real. Fibonacci! 🌀

New memoir working title: "No data, no problem: My life as a security researcher"

Exploring the iWill Covid vaccine risk simulator from Charité: iwill.charite.de Love the icon arrays to illustrate outcomes for vaccinated vs. unvaxx / 10,000, from side effects thru hospitalization to death. Love the personal/societal split levels. Lots to think about here for risk literacy.

Never fear, it's LOAD-BEARING turtles all the way down! Schölkopf & co (h/t @p-hunermund.com) on the dangers of simplifying for DAGs, Hardt & co on performative power & how classification IS causal modeling, & a cool study on how much scaffolding matters. wildetruth.substack.com/p/scaffoldin...

In *Causality,* Judea Pearl argues that SEM's founders kept causal meaning in their heads, but didn't encode it explicitly in the notation — and later generations lost track of that causal intent and correct reading.

Just got Judea Pearl's new book, and it contains this unexpected gem of wisdom about graduation ceremonies as essential rituals to embed accomplishment in community, enabling meaningful application. Pearl confesses he skipped all three of his own graduation ceremonies and regretted it.

"part of the peculiar beauty of *human* excellence just *is* its vulnerability… Human excellence is... something whose very nature it is to be in need... (Martha Nussbaum, *The Fragility of Goodness*, p. 2)." toleratedindividuality.wordpress.com/wp-content/u...

#Statistics thought of the day: If you think you can find new disease subtypes by empirically clustering patients, think again: www.fharrell.com/post/cluster... #StatsSky

Wow this is a breakthrough in validation of predictive models that fully accounts for point estimates used in predictions being only point estimates. Great work @richarddriley.bsky.social and colleagues! #Statistics #StatsSky

This Is Just To Say i have seen the draft structural equations that are at one level and which i was thinking worked don't Forgive me the code ran so precise and so clean

New blog post about the age-period-cohort identification problem! In which, for the first time ever, I ask "What's the mechanism?" and also suggest that sometimes you may actually *not* be interested in causal inference. www.the100.ci/2026/02/13/o...

Pushed a new module to my equilibrium screening repo. I was brainstorming this in a Google Doc in 2023. Today it's running Python with tests passing. The answer is still that we can't solve the validation problem in security. But now that's a coded feature, not just a claim in prose.

Up there with Alan Turing responding to a U.S. conference invitation: " 'I would not like the journey, and I detest America.' " www.atlasobscura.com/articles/tur...

My chopping is like my science; it ain't perfect, but it's done!

In which the "human in the loop" is the problem?

57% of my pilot participants aced Bayesian stats problems. The lit's ceiling is 24%. Either I'm the greatest science educator alive, or everyone used LLMs. Silver lining: my crappy data suggest a measure to estimate how contaminated your own crappy data are! wildetruth.substack.com/p/lowered-ex...

When you see "Post too long for email" in your current top Substack draft and think "ok, I've done my job."

New preprint! So, what's a multiverse analysis good for anyway?> With @jessicahullman.bsky.social and @statmodeling.bsky.social juliarohrer.com/wp-content/u...

DAGs describe discrete snapshots. Most causal mechanisms of interest in science on humans and other living things operate continuously. The two may not correspond. Is it time to panic? wildetruth.substack.com/p/can-we-bel...

Anyone else notice the hot water seems to be out when you're too tired to take a shower

writing: this is fun editing: this is not fun checking proofs: this is terrible reading my published work within c.5 years: this is horrific beyond measure reading my work c.6-10 years later: who wrote this

Wow an important paper. And Sander Greenland has thought deeply about this and brings rigorous thinking to causal inference debates.

Why is there not a saying "the cold takes 30 [IQ] points," like "the camera adds 10 pounds"?