Mostly because collaborative writing is very convenient. But yes apart from that I very much prefer a proper editor.
An additional (but minor) point is that setting up LaTeX can be a PITA.

Probably. But that seems a bit niche to me. You could always create a new package with the template and then copy-paste your code into that, no? I can't really imagine what other operations BestieTemplates.jl could be doing that are 1) quite tedious manually 2) not very brittle. Should test probably

Haven't tried it. What deters me is the dependence on Python. Seems very heavy-weight for a "simple" task of setting up a new package. I have PkgTemplates in my global environment, which I would like to keep as simple as possible. The only times I had issues installing packages was with Python deps.

Also adventofcode.com can be a great way to learn a language. It provides a lot of well specific (and fun) toy problems and you can also find a lot of solutions by other people (e.g. on Reddit) to learn how they tackled the problem. Reading code from other, more experienced, people helps greatly imo

I think the issues in the article are not related to reproducibility at all. Julia is great for reproducibilty actually. A project's Manifest.toml completely specifies everything making then environment 100% reproducible with no effort. RNG seeds need to be and in most scenarios can be controlled.

I think this article might be a bit overly dramatic nowadays. Personally, I used Julia extensively for 4 years of my PhD by now and never encountered any "correctness" bug (well not in package code that is - I have for sure produced a fair share of bugs in my own code). So YMMV.

Was du beschreibst hat einen Namen: Stochastischer Terrorismus
hateaid.org/stochastisch...
Ich finde diese "rechten Shitstorms" sollten viel direkter als das benannt werden was sie im Prinzip sind: Terrorismus.