Post

I love pymatgen. I started using it and contributing to it in 2018, back when I was just 21. So grateful to be included in this retrospective, alongside the massive (and still growing) pymatgen community. #opensource #matsci ⚗️

Shout-out to @molecularxtal.bsky.social .social @andrewrosen.bsky.social n.bsky.social @kavanaghsean.bsky.social and the other leads (Matt Horton, Aaron Kaplan, Janosh Riebesell, and Haoyu Yang, all - I think - not on BlueSky) for their work wrangling authors and putting this together.

As dope as I think this paper and pymatgen are, and as glad as I am to be recognized for my contributions, I must challenge the paper's frankly uncritical discussion of "LLM-based agents" and "AI-driven workflows", as well as its use of LLMs for analysis.

These choices, none necessary, none innocent, serve to legitimize a tool that is causing growing social, political, environmental, educational, and epistemological harm, that threatens the very foundations of science and research. #criticalAI

It has not been ignored. We added more discussion here, not as much as you wanted. However, I believe this paper is not the place for such a broad discussion on that topic. This paper represents 166 authors, from very different nations, with very different political systems.

I don't envy y'all trying to get >100 authors to agree on a manuscript. Truly, I see and appreciate the effort. That said, y'all did introduce the subject. You didn't have to mention LLMs at all (except perhaps re: CI). It seems inconsistent to say you can introduce it but not challenge it.

We challenged the maintainability aspect of adding LLM code but we can think about this during the revision as well.