Things that pass unnoticed, and the questions they open.
A tool, an experiment, a passage in a book, a puzzling study. Different starting points, the same habit: look closer, follow the sources, leave room for what I cannot explain.
After a motorcycle crash, a Tennessee music teacher began naming the chords he heard on the radio. He is now described as the first documented case of perfect pitch acquired through brain injury. The open question is what, exactly, was documented.
From every section, newest first.
Under propofol, the hippocampus of a few surgical patients kept tracking tones and words. The study calls that state “unconscious”: it is worth asking what the word rests on.
Fifty people who reported a near-death experience were asked to draw the spaces they remembered. Conical passages came out mostly without colour; open, arc-bounded spaces came out mostly chromatic.
On the verso of an authenticated Lorca manuscript, pencil lines begin «Canta el reloj». Whether he wrote them remains contested — and the speaker says he is not present.
In a hospice unit, researchers filmed people with advanced dementia for more than a thousand hours. Nine episodes of unexpected communication survived a multi-informant review — and only three of the twenty residents had any.
Hundreds of bottles emerge from a Swiss glacier. Their positions trace the movement of ice without preserving a perfect chronology.
Two researchers read the reasoning of the model that tops a leaderboard for AI agents. In almost a quarter of its attempts it cites, as instructions, rules that only the benchmark’s hidden tests contain. The study is quick, not peer reviewed and largely labelled by other AI models, and it cannot say how the model was trained.
Tools, models and technical curiosities, beyond the first impression.
Two researchers read the reasoning of the model that tops a leaderboard for AI agents. In almost a quarter of its attempts it cites, as instructions, rules that only the benchmark’s hidden tests contain. The study is quick, not peer reviewed and largely labelled by other AI models, and it cannot say how the model was trained.
Experiments I built to follow an idea. What worked, what surprised me, what remains unfinished.
I Wanted Something Small. The Web Went Back to the ’90s. Explore the section →Literature, cinema and philosophy: ideas about time, memory and identity, read as ideas rather than experimental claims.
I Am Not Here Explore the section →Phenomena at the limits of explanation. Each piece starts with a source or claim that can be checked, compared or challenged.
Without Consciousness, or Without Memory? Explore the section →I follow the source, check what I can, and leave what I do not know in view. There is no quota to fill.
Who writes here →