AI Cross (X) Course.

The classroom is changing. Sort of. Generative AI is now a co-student/textbook/professor/problem-solver to every student. What an excellent resource. It really is. But here's the thing. I want to evaluate the student, not its personal assistant. For that reason, the way we professors evaluate students must change

First, I will let you know what is safe, or more precisely, what need not change much. In-class exams are where it's at. The work on the paper is the work of the student, provided you are a sufficiently vigilant proctor. In the context of the mastery-based system I use, in which students are evaluated on their mastery of topics, rather than on their total scores on individual exams, and in which retakes of topics is possible, in-class assessments are incredibly valuable. I get to apply the time-pressure and the limited-resources-pressure (those resources being the student's brain and a formula sheet that I provide) that box students into performing to their best of their abilities. And further, with the option of retakes, students have the option to correct their learning and selectively-concentrate their studying. (And don't worry: retakes do not trivialize exams, as my students can attest. The retakes are sufficiently different from the originals, containing totally different problems, just divided into the same set of topics.) All-in-all, this examination system helps students master their learning (at least where the definition of mastery is attuned to match what we would expect of an undergraduate taking the course for the first time.)

And guess what, AI is a net positive in its interaction with this system. Students are provided a large set of solved practice problems, and AI helps them organize their study materials, unwind solutions, and unveil connections between topics. I've seen the materials my students have assembled using generative AI. And I have seen them ask questions live of AI in order to better understand a topic. The responses are helpful. However, an expert (e.g. mine or the TA's) perspective really enhances the responses. I have seen students prompt AI, only to receive a solution that they simply could not parse, or might not recognize as quite unusual (note: AI is usually correct, but sometimes its approaches are strange, such that a reader of the response may miss valuable insights). Nonetheless, I do believe having the response is better than having no response. And if a student is an excellent inquirer, they can interrogate the AI further to enhance the response. In my opinion, it is this tendency for incisive inquiry that distinguishes students who find success learning with AI from those do not find success. Q&A is exacly why I love using AI: I am replete with questions when learning a topic or a language, and I can just keep asking. So, in a sense, by mastering the incorporation of AI into studying, the student is mastering the art of inquiry. This is perhaps one of the most valuable skills we wish students to draw from their time at university.

So far so good. Now here is where things get rough. Homework is a disaster. Literally a disaster. To the extent that I am planning to cut it from the curriculum. I would estimate that virtually every student uses AI to assist in completing the homework. This use ranges from using AI to check their answers and provide hints, to using AI to generate answers that they may simply transcribe onto paper, or worse simply copy-paste into LaTeX. The issue is: I have no definitive way to assess which category a given homework submission falls in. I can look at an assignment (especially in the context of the student's broader performance in the course) and gauge that, yes, this student almost definitely used AI in generating their answer (like I said: the approach AI takes can sometimes be strange, and is always organized in a somewhat bullet-pointed fashion—basically very atypical of the natural productions of the human brain). But, there is no way for me to definitively show that the student used AI. It is math after all: an answer is either right or wrong. And so, what happens, is that an unknown proportion of students is submitting AI generated assignments, and my TAs are grading those AI generators, not the students themselves. That's messed up, and a waste of everyone's time. The students don't learn, and the TA's are grading ChatGPT. So, yes, homework is no good.

But then, a professor might ask, if you do not assign homework, how will students learn the content at home. Well, let me share an observation. I mentioned that I share practice problems with students for the exam. These practice problems are not actually new problems, but are really made up of homework and discussion problems. One time, I forgot to include a certain type of problem in the practice problems, BUT that type of problem was in the homework. When it came time for the exam, I included this type of problem and, guess what, the students struggled significantly more then they should have (this type of problem was not particularly difficult). What did this suggest to me: that they do not learn well from doing the homework, but they do learn from studying the practice problems. So, how will students learn the content at home? They will master the practice problems because they know the practice problems are essential for the exam, and they know they cannot bring AI into the exam. Therefore: my approach is to kill homework, but maintain practice problems, and simply compose a list of suggested practice problems after each lecture, for those students who wish to keep up with the material as it arrives.

Still, there are some components to the course that seem to be missing if we kill homework. Namely: how am I requiring students to do math each week? This is exacerbated by the great dip in student attendance in the wake of Covid. My attendance is "good" as far as many classes go (where attendance is not required, and note scaffolds are provided). But (towards the end of the semester) an attendance of 50% is a good day. So, the students who do not show up to the classroom, if they need not do homework, may go weeks without really thinking about math.

My plan is twofold. I thought to myself: in an era where AI will be able to carry out the math that most professionals will need (to be fair, even pre-AI very few professionals are actually doing math themselves), what skills can I help students improve, that will be valuable despite this. And I thought, based on my interactions with colleagues (what do I personly like and dislike), that what I value is showing-up-ness. Now, for me, showing up is not the state of being pressent, of being marked off on an attendance list. No, showing up is being there physically and mentally, using your mind to actively engage with the discussion. So: instead of usual attendance, I plan to utilize a system like the clickers of old, wherein students are required to answer brief multiple-choice questions about the lecture content, as it comes. In addition, I plan to require the classroom to be AI-free so that students rely on their brains alone (AI at home is fine, though). I currently give students several opportunities to try problems during class, but I have just never made a move to evaluate their attempts thus far. Now, those attempts will be evaluated, and form a part of their final grade, maybe 5-10%. Sounds scary, as answering questions correctly at the moment you are first learning them is challenging. Thus, each answer would be awarded perhaps 75% merely for an attempt (this is the attendance part), and the remaining 25% for correctness. Sounds less scary, I hope.

And then there is discussion section, which most math classes at USC hold. Again, since Covid, discussion section has across the board became a disaster when it comes to attendance. Near the end of the semester, my TAs report maybe 10% attendance. I do not blame the TAs. But, something must change. And so, again I though of a valuable skill I would like students to develop. In this case: collaboration. Moreso because collaboration is one of my personal weak points (I very much like to work alone, but really in part due to a kind of fear, that it would have been great if I had been taught to overcome). Therefore: my plan is for discussion to be turned into graded group work sessions. At the start of each session, students would be divided into groups and assigned a set of problems, and each group would submit exactly one set of solutions, composed by a pre-selected transcriber. My hope is that by randomly selecting a lone transcriber, students will be pressured to work with one another rather than work alone. Oh, and again, yes, these sessions are to be AI free (recall: at-home okay, in-class let's release ourselves from the bonds of AI).

I am hopeful that these shifts, of eliminating homework while maintaining practice problems, of including graded multiple-choice questions in the midst of AI-barred lectures, and of turning discussion section into AI-barred graded group work, will address some of the concerns that AI has exposed. To be honest: I no longer fear the rise of AI as it fits into the classroom. Actually: I think this generation of college students will have the most problems using it responsibly and effectively, because they have not been taught how to do so. Future generations will surely arrive at university far more savvy in its effective use. And, I think the changes that AI will force instructors to make will ultimately improve the quality of educational approaches in university-level mathematics, which have remained largely stagnant.

date: 2026 Thu May 28

Creature of Habit: 7–Eleven.

One thing you can most certainly accuse me of is being boring. Or to put it more precisely, for being consistent to that point of being boring. Once I settle into a routine, a significant external force is required to extract me from it. In early grad school this manifested itself in the literal identical set of three meals I would prepare each day, only occasionally broken by a rare outing. To many this is a sign that the first circle of hell has been entered. To me this is joy. This is not to say I am a picky eater. Au contraire, I can eat any genre of food and be content. Rather, I, not unlike the flow of electrons, am drawn to the path of least reistance. For me, that path is the path of constancy. Of course, now, being a member of a family at home, and not wishing to subject said family members to my peculiar tendencies, my diet has become more varied, at least when I am eating alongside them (though now, our discrepancy in preferences rarely arises, because my mother–in–law does all the cooking. I know, lucky fella I am).

But in those meals that apply only to me, I revert to my primitive ways. This devolution is most striking in my daily relations with 7–Eleven. As luck (or misfortune) would have it, there is a location within walking distance of my home. And so, of course, it has become a routine for me to visit each morning, so that, as a subsitute for a healthy breakfast, I can enjoy an unhealthy one. My beverage of choice is a large Monster Zero–okay, look: I have made strides here, as I used to depend on the full–sugar versions, until this dependency, and aging, led to my sugar intake outpacing my metabolism. There is something about the bliss of a sip from this divine beverage in the morning that, well, reminds me that I am fake–sugar and caffeine addict and should probably address that.

And for my meal? Only the finest, freshest, set of buffalo chicken rollers (or on occassion, chicken taquitos) aged to perfection on rollers (is that the reason for the name?!) under a heat lamp. You can really tell, in each delectable bite, how the hundreds of chickens, whose least valuable parts were ground into this fine tubular form, lived. (Hint: they did not live well.) Well, I should not mislead, as it is not every day that the chicken roller is aged to perfection. It is rather like playing the lottery (isn't this fun?). If you are lucky and they are fresh (that means freshly prepared in the back, right?) then the texture is just right and I have entered heaven. But on some days, when things aren't going your way, you will encounter what seems to be a sort of dried husk of the original. Even on these days, though, you remain grateful, and partake anways, hoping that the next day will bring greater fortune. And if you are supremely unlucky, your routine could be spoiled by an absence of the item you desire. On these days, with a sigh, you reach for your second, or even third choice, and pray to the 7–Eleven gods (whose business hours unfortunately match the store's moniker) for a greater bounty tomorrow.

Now, I can't blame this on the birth of my daughter, but she certainly didn't help. By this I mean that, prior to Eliana, there was some modicum of moderation. On days I did not go to campus, I would prepare breakfast at home for me and my wife, rather like a sensible adult. But, once Eliana was born, she creeped into my morning routine. My trip to 7–Eleven doubled as a walk for my daughter, in preparation for her morning nap. And so, in an appaling abuse of logic, I determined that my visit to 7–Eleven was required as an essential part of not just my, but my daughter's routine. There was some truth to this: Eliana would sleep quite soundly after this morning walk. Though she no longer regularly needs a morning nap, she still comes strapped with me to 7–Eleven each day that I am not headed to campus. She is a popular item with the employees, who know her by name, and are always trying to coax out a baby smile. They have (half in jest?) even suggested picking up a dress from their native Ethiopia for her, and asserted that they'll babysit her one day. So, if you think about it, how can I decline this trip to 7–Eleven, when so many people have become wrapped up in it. Now, you might say, why not just get the drink and skip the questionable taquitos. To that I say, please stop making suggestions, you are upsetting the baby.

date: 2026 Wed May 13

Cantonese.

Cantonese is perhaps my greatest current challenge. The pressure is mounting with my daughter Eliana reaching those early years where vocabulary has begun creeping into her mind. My wife's first language is Cantonese. My mother–in–law currently resides with us and near–exclusively communicates in Cantonese. And naturally we wish to preserve the language in the vessel of our sweet daughter, Eliana. Especially with its gradual fading away in mainland China as Mandarin forcibly reigns supreme in early education.

Now it is not as if I have simply not set aside time to learn the language. Oh to the contrary I have spent years practicing. I have gone through textbooks and grammars and have assembled flashcards via Anki that I practice daily. I have watched children's shows (Peppa Pig!) and read children's books (Hahm Baahng Laahng). The language is spoken everday at home. And I overhear it thrice a week when I venture into San Gabriel Valley to play badminton. A popular pastime among the Cantonese diaspora.

I can more or less understand the Cantonese spoken in the household. My mother–in–law and I have settled in an equilibrium wherein she will speak in Cantonese to me and I will respond primarily in English, and vice versa. Our comprehension of each other's native languages is sufficient that this state of relations is stable. Eliana, my daughter, mainly understands Cantonese words, and so it is in Cantonese that I will communicate commands or remarks with her. I will also use Chinglish, a mishmash of Chinese and English. In a sense: this is progress!

But here is where my limits expose themselves. Of course there is the written aspect: I have made no attempt to learn the written language as my intention is verbal communication. Writing and reading constitute a whole gargantuation task of their own. For verbal communication my two greatest weaknesses are: listening comprehension, and the mustering of courage to use the language.

Despite hearing Cantonese daily (spoken in my vicinity and through audio flashcards) my ability to accurately distinguish tones is poor. I would not say the skill is null: occassionally I surprise myself. But I rely heavily on context. If a series of words were spoken to me in succession but without any particular meaning, I would falter with identifying their tones. Compare this with my wife who in an instant can identify a tone that is off, perhaps when I mistakenly use a mid–level tone instead of a mid–rising. I fear this skill will be the most difficult to master as I was never required to call upon it in my developmental years. For Eliana, she will certainly receive a leg up here! Nonetheless I do not intend to give up. This summer I plan to schedule more concentrated viewings of Cantonese (children's) television. The dedicated exposure I hope will make a difference.

The part of my mind that allows me to excel in the deep–thinking required of abstract mathematics faces a counter in spoken languages. My thinking is highly detail–oriented and slow–paced. I am the one who fails to see the forest for the trees. In mathematics this has advantages as I can catch connections that evade others. In spoken language, however, there is no time (at least for me) to reflect and consider carefully each word I hear or intend to say. But my brain is stubborn and tries to grab details. Invariably this leads to a missing of the core of what was said, or to fail to compose a sentence due to a focused uncertainty about the correctness of a particular aspect of its form. In the end what needs to be done is to let go. An allowance of the language to wash over me without an obsession with catching each particle. My wife is skilled at seeing the forest despite the trees. And I believe most who are able to learn languages flently later in life fall into this category too.

And my greatest flaw: fear. Yes. So much about mastering language is finding the courage to actually use it. In my work life: I love planning and coming in prepared. But conversation is improvisational. And when you improvise you will make mistakes. There will be instances when your conversation partner's face scrunches up as they struggle to understand what you just said. For some (my mother first into his category and is quite successful with foreign languages as a result) it is easy to ignore this moment of awkwardness and power through or try again. For me that moment triggers a fear response. Totally irrational and honestly silly but this is my body's default. This is only made more ridiculous by the fact that Cantonese people are so kind and supportive when you try to speak their language. I suspect some of them realize the difficulty their language poses and so are generous with giving grace. So it is my personal mindset which punishes me here. My obsession with perfection on the first attempt. No external factor is to blame. I know what to do though. You don't have to say it. I just have to use the language more, have more of these mini–failures, get comfortable with them, see them as the harmless components of social interaction that they are. Perhaps writing this blog post will instill me with greater motivation to put myself out there.

So yes Cantonese is an incredible challenge for me. But it is also a joyful challenge. It is a beautiful language which is incredibly sensible: unlike English, it is a fundamentally pure language, so the rules of its grammar are simple and sensible. It is purer in the historical sense than Mandarin even. But it is also expressive. Do not even get me started on the sentence–ending particles, which endow phrases with texture, and which I have the barest comfort level with—their mastery is another task whose completion is distant. I truly enjoy the language and am thrilled that Eliana will learn it alongside me.

date: 2026 Tue May 12