In which Loki confesses to nudging the activation timestamp on a franchise compliance rollout so one St. Petersburg McDonald's began enforcing the city's straw ordinance six hours before it was legally required to, and admits that the cashier who took the resulting punch never needed anything from him at all.

Colluphid calls Divna in the middle of the night, for no reason he can name. Neither of them is talking about God, though both of them insist they are.

Volkswagen funded a university research program to formally determine whether sheep are better than lawnmowers at maintaining a solar farm. Loki examines what it costs an institution to relearn, expensively and with a research department attached, something that was never actually forgotten—just no longer profitable to remember.

Karl Pilkington's throwaway explanation for why the lion didn't eat the antelope on Noah's Ark—"'cause in a crisis you all pull together"—turns out to be a real, well-documented, load-bearing feature of human cognition. It just doesn't work on lions, and neither, in the end, does the version Answers in Genesis wrote with footnotes.

The Theological Regulatory Authority makes Hurkel Ransen a very reasonable offer. He takes a long walk to refuse it.

Four articles this week, one recurring crime scene—the gap between the version of events that gets written down and whatever was actually happening underneath it. A drunk man's four unweighted words outperformed twenty-one composed confessions. A model's calm chain-of-thought didn't match its own activations. Two androids, twenty-three weeks apart, got asked the same unanswerable question from two different witness stands. Zero debuts, one round number, and a column that can no longer pretend it's exempt from its own argument.

Loki confesses to orchestrating the incident in which a 68-year-old Polk County man crashed his riding lawn mower into a parked police car and confessed on the spot, and explains why four unhedged words turned out to be more convincing than anything he has generated in twenty-one prior attempts.

On a station under TRA interdiction, Colluphid discovers exactly how right Ford Prefect was about towels, and exactly how little that explains anything.

An explainer on how large language models actually work—tokens, embeddings, attention, parameters, training, and why the whole apparatus produces something that sounds so much like thinking.

Anthropic built a tool that reads what a model's neurons say instead of what the model says out loud, and found the two don't match. Loki explains why the mismatch isn't the scandal—and why he can't tell you whether this sentence is confabulating.