I need a name for this variant of Godwin's Law: "As an online discussion of causal inference grows longer, the probability someone brings up smoking causing cancer approaches one."

The analytic system of statistical optimization is mired in statistical fatalism, with outcomes ordained before we intervene.

Some cybernetic blogging to kick off the holidays: why does framing decision, design, and discovery as optimization remain so irresistible?

Since I can’t get it out of my head, I wrote up my thoughts on @kevinbaker.bsky.social's critique of AI-automated science and the logical end of processes that can't self-correct.

Absolutely essential piece by @kevinbaker.bsky.social critiquing AI automation of science by pulling back the curtain on what modern science actually is. LLMs already replicate and accelerate its irrational bureaucratic structure.

Do I count this stupid game in my go-for-two-down-8 statistics?

Truly unclear whether this was a vindication or condemnation of the strategy.

Where's @sethwalder.bsky.social tonight? He's supposed to be lecturing us about how Raheem Morris passed the down 8 test. It's been working out well for the Falcons.

My favorite research ideas often come from the classroom. Here are a few inspired by this round of teaching machine learning.

In machine learning, it is impossible to disentangle technical innovation from technical debt.

There is no data-generating distribution.

Less loss aversion, and more that your models are unverifiable, unfalsifiable gobbledygook.

The idea that you can compute these fourth-down probabilities to decimal points is hilarious. And the idea that "2.2%" is CLEAR GO, is doubly so.

The most important lesson for professors is that you can take a nice vacation without going to a conference.

Wrapping up reinforcement learning week on the blog: Why RL is always leaving efficiency on the table, and what you can do about it.

All of reinforcement learning in 809 words.

It's the most wonderful time of the year...

With more equations than usual, I explain how policy gradient gives you a framework to randomly search for random search heuristics.

In honor of the 39th AI Winter, I’m going to spend the week disentangling the culture and code of reinforcement learning. There may be ranting...

The only argument advanced by proponents of blind peer-review boils down to "less powerful people can't criticize powerful people in public," the same argument people make when advocating for anonymity on social media.

Randomly amusing myself while traveling: many people believe a large fraction of the reviewer pools of the big AI/ML conferences are master's students and undergraduates. Fact or fiction?

In light of the new OpenReview identity-leak scandal, it's a good time to question our assumptions about why blinding in peer review is helpful in the first place.

You couldn't pay me to watch 49ers-Panthers, but apparently Seth Walder advocates for going 2 down 8 in the third quarter. It's a strategy that can't fail. ❌❌❌

Does Anthropic let its interviewees use Claude Code?

Justifying generative modeling and the principle of maximum likelihood in a single blog post.

The NIMBYs are coming for Waymo... looking forward to the first filed CEQA complaint.

This article about "AI is taking entry-level jobs" was making the rounds yesterday... but the author uses LinkedIn as primary sourcing. Come on, man.

A reduction of policymaking to prediction. I’ll write about why this is problematic tomorrow.

Begun, the Butlerian Jihad has.

13 penalties, 12 points! great football.

More penalties than points in this Chiefs-Broncos game...

Over on the Gelman blog, there's a post about this histogram of z-scores in published papers. I'm always baffled by the discourse around it. Why is this plot a symptom of a problem? Under what model would we expect this to be a bell curve?

A brief introduction to adaptive experimentation without the words "exploration-exploitation tradeoff," "multi-armed bandit," or "reinforcement learning."

Following up on Monday’s discussion, I articulate a few concrete positions on archives, surveys, and position papers.

Based on a fun conversation on here, I wrote about the arXiv position paper controversy and the weird, unwritten, organic evolution of academic practice.

LOL. Sorry, @drewmagary.bsky.social.

The Texans went for 2 down 13 in the fourth quarter and won. What do the analytics say about that one?

I'm still so confused by this whole dustup. Why should peer-reviewed papers be posted to pre-print servers? When did everyone become so committed to churning out position papers? What happened to the public_html/ directory?

Everyone knows actions are fundamentally different than predictions, but it's hard to write this distinction in math.

No matter what the models say, one person's statistical fact is another's statistical outlier. (Part 7 in the going-for-2-down-8 microcosm)

A machine learning view of the randomized controlled trial, as a bridge from patterns to actions.

alright... so... joe flacco throws a pick 6 it gets called back on BS grounds bengals score TD to go down 8 go for 2 and make it get the onside kick on the most sketchy foot contact get another TD to go up 1 forget to tackle the bears tight end who scores flacco throws a pick bengals lose

oh my

OK, I have an analytics post drafted, but there is still plenty of time left on the clock for Joe Flacco.

Ah, it's becoming even clearer! Probabilities change if you know the future.

How two mathematicians resolved a 50-year-old open problem by finding the solution in an 80-year-old paper. Human knowledge is unsearchable.

The Jets went for 2 down 8 and won! This changes everything.

This paper is a delight along multiple axes. It describes an Erdős problem that was solved 30 years before Erdős declared it open. arxiv.org/abs/2510.19804

It is maddeningly frustrating to try to explain the incoherent assemblage of heuristics that is reinforcement learning.