Post #2885139

@lina@vt.social @ahltorp@mastodon.nu OT but I really miss .map files that told you the type and name of every variable in a routine. I work with a lot of legacy scientific codes (you'd probably be appalled at the amount of F66 still out there) and just having that inventory of variables helps immensely for documenting the code and understanding it. I work with accident analysis codes and we have enough trouble benchmarking them against experiment that we don't need unforced errors like LLM slop making things worse. But we shouldn't accept errors by cheap stochastic automation when thoughtful dependable deterministic techniques are available regardless of what kind of code we build. I wish more devs were in my position with a work culture and regulatory framework that puts safety first, encourages problem reporting, and has actual consequences for people subverting the process to ship garbage. Not perfect but it's nice being able to point to federal cases where people have literally been banned from doing this work because they cannot be trusted. https://www.nrc.gov/reading-rm/doc-collections/enforcement/actions/individuals/index