Reproducing scientific papers - with Rust & "AI"

"We do not share the code, but all information is in the paper anyway."

How often have you been told this bizarre sentence or read it somewhere? Times change and with a solid collection of literature, a few Rust crates covering the core aspects of a subject and of cause come patience, tools like Claude are opening really interesting new avenues.

My personal "first case study": I threw a paper about astrodynamics, how objects move in certain ways through space, at Claude - I had been interested in reproducing it for many years, though the authors rejected to share code again and again. I had not only this one paper in my collection but just about every citation in both directions, i.e. both papers my paper of interest was citing and papers citing my paper of interest in follow-ups. And I had a shiny new high-performance Rust crate offering a lot of astrodynamics "primitives", building blocks for more complex simulations, plus of cause some cloud credits to spare for raw compute power. What would have taken me probably a couple of months with "good old manual labor", collapsed to 48 hours of "supervised work" and a net investment of around 50 Euro.

This is not about praising AI tools. It is a testament to how far the open source ecosystem specifically for Rust has come combined with the fact that it has become feasible to automatically scan large bodies of texts and generate matching code. The results are by no means perfect, but they are really good starting points for understanding, learning and further investigation. Just about every scientific publication that is not entirely made up is fair game now and this is saying something. Use your imagination what publication you would really like to understand and play with and what you could possibly do with the results, both personally but also professionally.

-Sebastian