2024 - 11 November

The SLS is still a national disgrace

A rough read as a current NASA employee. One wonders if this is already circulating around the DOGE. We’ll see next year.

Matrix Calculus for Machine Learning and Beyond

This is a short course from MIT Open Courseware, with all lectures online. The goal is to fill the pedagogical gap around vector and matrix derivatives, concepts that are required for nearly all higher level applied math (e.g. all of engineering, statistics, machine learning).

I’m not claiming to be experience enough with teaching calculus 1 to know that we could start like this, I found the reframe of a derivative as the linear operator $f’$ which maps small changes in the input, $dx$, to small output changes $df \triangleq f(x + dx) - f(x)$ (i.e. $df = f’(x)dx$)

I think the first 4 lectures are extremely useful (with the exception of the Kronecker product lecture, which was interesting but possibly overly complicated and less used), and the last four could be skimmed as needed. The second half was spiky

How Did You Do On The AI Art Turing Test?

50 questions guessing “Made by Human? or AI?” People who took the quiz (i.e. ACX subscribers) got it right only about 60% of the time. Interesting points:

• 5 people were able to get 49/50 correct.
• The article made a lot of rounds on the internet, and there are many rebuttals and defenses that seem to be starting from completely different positions than Scott Alexander does.
• Many people are using it as a placehold for “Does X count as art?”

What role does hydrological science play in the age of machine learning?

Apparently the title question is an inversion on one asked by a hydrologist in 2019. This article is simultaneously too dry for most people, while also written in a style that’s more provocative overtly opinionated than 99% of academic articles. Not having a horse in the hydrological race, I very much enjoyed it.