I'm reflecting on the end of the quarter, and a bit on the end of my first year.
This quarter, I formally took two independent studies and a doctoral writing class, and was the TA for an undergraduate project class. The doctoral writing class was not valuable for me. I don't feel like I contributed very much to the work other students presented, and fairly, I didn't gain much either. I don't know if a workshop format works for me. I haven't yet found a strong format for collaborative writing.
Strangely, the TA experience was really valuable, as it gave me a view of how AI is being used in project work. The class is the capstone project for all seniors in computer science and informatics, and my section had about 60 students in it. Previously, they had worked for 10 weeks on the concepting phases, and this quarter was dedicated to production. The output was to be a working prototype.
I observed that they were able to make systems that were extremely functional, and to do it incredibly quickly. One team decided to add a phone-based interface, where the system automatically calls a user-entered number, the recipient of the call speaks, the system schedules an appointment, and the call is recorded and transcribed. They implemented this, fully, in a week. Another team developed a feature where a user could draw a curved line on their phone, and the line would become a roller coaster with an animated figure riding it. Another team made an app that took a photo of a user's face, analyzed their makeup, and gave them suggestions on how to improve. Nearly all of the teams were using some sort of machine learning and LLM in their products, and had these capabilities working fairly well.
Unsurprisingly, much of the work was not refined (it's not supposed to be—that’s the whole point of school), but what I wasn't prepared for is how much AI slop was in it, both in aesthetics and product. With few exceptions, most of the design was driven by what the AI proposed, not what the students generally wanted. Several teams had pretty well thought-out design ideas drawn in Figma, but others were typing what they wanted into Claude and then dropping the result into their work. Claude likes to helpfully add random UI noise and so what could have been a simple interface becomes crowded with buttons and glyphs and social tools and graphs and charts that serve no purpose. It's clearly recommending the addition of features that have no business being in the core product, and again unsurprisingly, the students follow the suggestions. They aren’t questioning product value, and so features that make no sense end up in their results.
Surprising to me is how much generalized understanding of software production the students had learned, yet how little understanding of their own decision-making that they displayed. They spoke of deployment pipelines, restful endpoints, CORS errors, merge conflicts, cloud storage, model training, standups, test cases, and more, and they spoke with more clarity than I would have expected a student to have. I think their language was not just empty—that most of them really had a mental model of how these different components were working together at a structural level. But when asked about details, they largely had no idea what was happening. I asked simple questions, like "did you use a package for that or did you write it yourself?" and the answer would be "I used Claude." or, "where will you persist that user-generated content?" and the answer was "I'll ask Claude."
I was expecting to see dependency on AI for syntax and code writing, but there appears to be dependency on more fundamental decision making and questioning; there seemed to be no skepticism as to what was being presented by Claude. The code worked, so it must be good. I don't know if this is a net positive or negative compared to "older" approaches to education; in the past, even getting something to function at all was the challenge. I think these students are probably prepared for junior engineering roles, but their work product will be garbage and they will need a lot of pretty direct guidance, real-time mentorship, and lots and lots of code reviews. I don’t know if companies have this sort of job anymore, and I can’t imagine that many companies would be willing to offer that sort of time commitment to quality control (although they certainly should!)
On my own work, I made a lot of progress, and I'm proud of it.
I worked through the rest of my literature review reading list, and wrote my comps paper. There are a few sections I'm really pleased with, particularly one focused on chronopolitics and one focused on risk. Both will end up integrated into some future paper or idea.
I presented my first paper at CHI, and learned a lot about how that conference has changed since I used to go in the early 2000s.
I revised my faculty research paper, and resubmitted that.
I started and completed my student research, (I'm behind on writing about it here...) and I'm in the process of analyzing that data and working on a summary of it. I think I've gotten my head around discourse analysis, and really appreciate it as an approach; I'm using that to interpret content from this research. The results are fascinating, and I'll have a draft shortly.
And, I think I have a high-level plan for the next year. I'm targeting late Fall of 2027 for a defense and completion.
I'm enjoying this process a great deal, and so far, the experience has been what I hoped for.