Jonathan Aldrich

Professor of Programming Languages and Software Engineering at CMU. ACM Publications Board member. He/him. Views are mine.

Last Notes

2026-05-21 16:38:17 GMT

@nprofile…curl decided a while back not to license papers in ACM's digital library for LLM training--because of concerns about mischaracterizing our research and whether the community wants its papers used this way.
This is now being discussed again and the decision may change.

2025-12-20 02:17:39 GMT

When there's a word for something, use it, rather than a negation. So, prefer "forbid" to "disallow." Prefer "range" to "codomain."

2025-11-20 01:36:56 GMT

Need HotCRP bidding language. Example program:
bid=-99 where "LLM" in abstract

2025-05-24 14:45:04 GMT

- reply

@nprofile…6kk0 Thanks for the explanation. It does seem like better tooling could help both (1) and (2), actually! But these are both hard problems, (1) in terms of a wide interface to system settings and (2) in terms of testing updates (as @nprofile…2key describes). One could imagine at least using shared libraries (of whatever version the developer requests) for (2) instead of replicating for each app--scales better.

2025-05-24 12:28:22 GMT

- reply

@nprofile…6kk0 this is the first I've heard of flatpak. I looked it up (https://flatpak.org/) and the advantages listed seem compelling...seems like an obvious win. Are these not actually good reasons for flatpak? How does it cause security and customization problems?

2025-05-20 16:20:27 GMT

- reply

@nprofile…6kk0 I can't quantify the damage done by silently broken expectations. But I can say that the types in TypeScript are a huge productivity win for me vs. programming in JS and having no idea what type anything is.
Admittedly, I am one of those who understands the limitations of TypeScript. They occasionally cause me problems, but it's surprisingly rare.

2025-05-20 16:13:22 GMT

- reply

@nprofile…6kk0 I think your observations are right. However, having started to learn programming before Java, and having programmed in TypeScript, these are both fantastic and influential advances over their mainstream predecessors! Java provided a safe, GC'd language that was adoptable in industry and (comparably) easy to learn. TypeScript leaves many gaps (and as you say may provide misleading expectations) but also introduces people to types and aids productivity vs. JavaScript.

2025-04-29 02:12:50 GMT

The latest human subjects research scandal: the University of Zurich research group experiments on Reddit users w/o consent.
Just don't do it, folks....
https://www.reddit.com/r/changemyview/comments/1k8b2hj/meta_unauthorized_experiment_on_cmv_involving/

2025-03-12 12:37:52 GMT

- reply

@nprofile…9374 @nprofile…6kk0 @nprofile…avjd all great points! I would add that "double-dipping" is explicitly accepted as OK when you do conference -> journal. You have to add >30% or so new material, but everyone knows the other 60-70% is reused and we're OK with it because the journal is the definitive version.
Double dipping for a published workshop -> conference is OK for the same reason.

2025-03-06 17:37:08 GMT

The one situation where double-anonymous peer review gets really awkward is when you have published a workshop paper and now you are submitting the full version to a conference. Not citing the workshop paper causes problems, but citing it in the third person devalues your work.

2024-07-19 23:28:30 GMT

Calling Mastodon security researchers: what's the *right* solution here--the one we should be developing/using instead of CrowdStrike?