During the week, I often collect and share links I find interesting on Elder Research’s Slack. I shared a short list on LinkedIn and reproduce it here.

A short link roundup for this week:

1. More writing about LLMs and programming. There are some real long-term concerns around ‘LLM dependency,’ but they do seem to provide a real boost to more senior folks. Matt Bezdek points out, too, that good StackOverflow answers come from folks with deep subject knowledge, which is something LLMs just don’t have (yet?); the ability to edit and augment generated code is super important.

   “New Junior Developers Can’t Actually Code
   “The 70% problem: Hard truths about AI-assisted coding”

2. Gaussian processes vs. FFTs or Fourier terms in models. GPs are expensive (though Hilbert space GPs are better), but they fit data in a more natural space (time, not frequency).

   “Modeling Autocorrelation: FFT vs Gaussian Processes”
   “State Space Models and Structural Time Series, with Jesse Grabowski”

3. A neat idea for a prior (ARR2) that conditions time-series models based on expected R-squared (e.g., neither 0 nor 1) and makes models more robust.

   “The ARR2 Prior” on arXiv

4. A review of the excellent uv tool set for Python, one year on. (Spoiler: It’s a really good tool.)

   “A year of uv: pros, cons, and should you migrate”

5. A gallery of figures with accompanying code for producing them using base R graphics. (I like ggplot2 just fine but would like to use the base system more.)

   “basegraphics”

This post is kindly republished by R-bloggers.