You definitely can use LLMs to do your modeling. But sometimes you need very fast, cheap, and smaller models instead. Also there's research out there showing that using LLM to generate training data for targeted & specific models may result in better performance.
I'm very happy to see "The Breeder's Equation" on Hacker News. Wow. I spent my graduate career studying mathematical biology and working with biologists.
It's a fascinating equation and I spent some time on it converting it from a discrete equation into a continuous equation along with explaining the use cases and motivations behind it here:
http://fromdata.org/2013/10/27/the-continuous-breeders-equat...
This of course, does not take into account mutation, transduction, conjugation and other sources of genetic variation.
Edit: Please take these approximate expectation equations with a grain of salt. They only apply to larger, discrete-reproducing, controlled populations. Applying results like these on spatially segregated or small populations does not really work. Also this equation does break down a bit when the phenotypic trait is a result of very complicated interactions between genetics and the environment.
