Field notes from the web

This is mostly academic chit-chatter, but I enjoyed reading about potential advantages of slowly (but continuously) improving a course.

A short article about the idea of semantic ablation, which is the observation that A.I. writing tends to be rather uniform. (Side note: I do not know what the register is…).

Articles about Japanese explainer books with beautiful dynamic illustrations.

Fun fact: A flexible curve/Lesbian rule was a flexible mason’s rule used to measure and/or reproduce irregular curves.

Here is a very interesting article discussing the problems that appear when the digital A.I. world meets the real world.

Unsolved historical mysteries.

Science: This article discusses the notation of calibration for survey statistics (however, the content generally applies to calibration). Nice thinking.

Random videos: Here is a video about milk soup. And here is one about the joy of assassination movies. This video about the history of folding knives is just fascinating (and I am not even interested in knives).In this one Matty Matheson promotes his „new“ book from a year ago. Furthermore, this video provides a very, very short introduction to hyper connections; and this one claims that the dream of quantum computing is currently falling apart. There is also this, this, this, and this — all short and worth watching.

Journalism

Inspired by this article about economists as reporters I watched a lot of videos about (mostly investigative) journalism. Here is the current list:

Water

This article is written a bit strangle, but provides a nice conceptualization of “water bankrupcy“ and many reference for further reading.

Kraft et al. (2025) examine an interesting way for providing quantile predictions called. For me it remains unclear (in terms of mechanism) why it performs better than directly estimating the quantiles. Maybe the augmented batch leads to better generalization?

Probably not surprising, but different model structures lead to different generalization patterns — and this can be exploited (e.g., with ensembles). A particular example for hybrid models was now proposed by Song et al. (2026).

This paper by Astagneau et al. (2026) about a softer calibration objective looks interesting, but I. have to chew on it a bit longer.

Food

I got recommended this restaurant from Graz and this one from Vienna. However, the former is is Graz and the latter looks very, very expensive…