I'm finding old plans for software projects that never got built and I'm wondering if there is any research value in developing a full implementation using the languages of the day to finish the job. Could be informative and would produce something people could interact with, not just read about
Hunter
@hdemeyer.bsky.social
Informatics PhD student @ UIUC Software Studies & Technical Writing Doing qualitative research on Research Software Engineering Interested in tool development, type systems, rhetoric, version control, computer history, philosophy of science
"Writing to think" wrt to coding is one of the major themes I've seen in my conversations with RSEs, students and professionals, and it is why I don't think AI use should be condoned in educational settings. It is robbing students of thinking opportunities the same way it does when writing an essay
Will I use Claude going forward? Not yet. I cancelled my subscription. Writing (and rewriting) code by hand is how I think and make architecture design decisions. I don't want to offload that. Also, the environmental impact of data centers is definitely not worth what it provides me.
currently writing research software to study research software
This is the best kind of bike rack and all other kinds of bike rack are wrong
it took me over an hour to convert timestamps i hate computers
Did my first integration test last night of my bike lights project. Still need to get some acrylic laser-cut to tint the tail lights red but I'm pretty happy with the brightness/visibility. Next I'll be working on a more complex pair of retroreflectors for the front and a control interface
The what/how of provenance tracking in vibe coding is something we should be studying imo, probably longitudinally. *I* feel like it is/will be important, but I've heard a range of opinions from RSEs
adding ".vibe" before the main file extension like "file.vibe.js" to indicate i haven't read the code
I had to work with a W*rd docx file today to submit my manuscript and I forgot how awful WYSIWYG editors are. Idk how people write like that every day
if we called the command-line a chat interface would students be less opposed to learning how to use it
biking home from the beer garden while tornado sirens are blaring is peak midwest
Working through some Whisper transcriptions this week and the variance in how things are transcribed between different interviews is surprisingly wide. Capitalizations that exist in some files but not others, um's and uh's placed inconsistently, sentences of different lengths (or no periods at all)
listening to mission control make up data provenance procedures for moon images & notes on the fly and then hearing the astronauts describe their own, different procedures based on their experience in the environment is both fascinating and infuriating also can't believe they use onenote
nasa is really out of practice televising shuttle missions...the feed blacked out on launch and they didn't show booster separation ☹️
"AI is just another tool" No. Tools don't have logins, subscriptions, or silent updates that change how they function. You own tools. You control tools. AI is a *platform* until this is what coding with LLMs means (and if it becomes a tool we should remember that some tools require training to use)
like, the absolute minimum you would need to do to make ai generated code sustainable would be to transition to local open models. "everyone should use llms" is just absolutely not an acceptable option when it means "everyone should pay a small number of increasingly powerful and evil companies"
I tried to use the AI in colab and it fucking sucks it takes forever to do something, doesn't let me see or edit the code it generates before running, generates a bunch of text to describe what it thinks I want then inserts a bunch of md and can't fix its own error how do people work like this
I don't normally listen to podcasts but people have been sending them to me and I find it really odd that podcast people keep saying that generated code "either runs or it doesn't" and that is why AI works for code like...software has never worked that way? Some bugs are not obvious? Am I crazy?
unfortunate that PrairieLearn is snappiest on Friday night when demand is low but it makes my grading go a lot faster so ig I'm staying in
damn i really told a room full of social scientists that i use emacs
how long before vibe-coded software starts including ads in the form of comments or print statements or log messages
Remember: LLMs and bots and AI-generated things should always be called ‘it’. Regardless of the style or tone of voice, there is only an ‘it’ there. Don’t humanize it.
feeling growth as a grad student because I'm filling out IRB paperwork for the first time
still in shock that I got this book, which was published this year, for $6 in a coffee shop & used book store in Maryland this weekend
UIUC CS is offering at least two "applied AI" courses in the spring, targeted at non-majors. Adapting curricula in this way makes sense in the current moment, but a CS department is the last place I'd want to take this kind of course. Urban planners don't learn about vehicles in a MechE class. 1/
I'm editing a final paper for one of my classes and got stuck so many times on sentence-level edits. Yesterday I printed it and did read-only marking of changes I wanted to make and its helped me make progress but also gave me a better idea of structural changes that would enhance clarity
Sony is trying to use digital signatures and depth information from camera sensors to prove authenticity of images authenticity.sony.net/camera/en-us...
Camera Authenticity Solution | Sony
authenticity.sony.net