Wanted: Advice from CS teachers

@futurebird Do they work in groups? And ask them to consult with other groups . Record that plays "It isn't broken, the code is doing exactly what you asked it to do" on loop.

@futurebird

I’ve taught programming like this, but I’m an increasingly huge fan of the debugging-first approach that a few people have been trying more recently. In this model, you don’t teach people to write code first, you teach them to fix code first.

I’ve seen a bunch of variations of this. If you have some kind of IDE (Smalltalk is beautiful for this, but other languages usually have the minimum requirements) then you can start with some working code and have them single-step through it and inspect variables to see if the behaviour reflects their intuition. Then you can give them nearly correct code and have them use that tool to fix the issues.

Only once they’re comfortable with that do you have them start writing code.

Otherwise it’s like teaching them to write an essay without first teaching them how to erase and redraft. If you teach people to get stuck before teaching them how to unstick themselves, it’s not surprising that they stop and give up at that point.

@ligasser @futurebird I'm stealing this just so I can start saying, "You're interrupting my me time." Thanks.

@david_chisnall

Tangentially related:

"AI can write code so why teach how to code?"

"Great point! It can write an essay too, so why teach how to read."

Like. We've had calculators for decades and still teach arithmetic. And functionally the average person needs to know probably more about mathematics and needs to read more than they did a century ago. The same will apply for code.

@futurebird @david_chisnall I mean… if AI could do what it promises, why are these companies hiring?

@david_chisnall @futurebird it's not exactly the same, but it feels like the same ballpark ... When I started learning to program, the instructor used a pseudo language (based on Pascal, but modified) so there was no compiler available.
We analysed problem code visually before writing our own, but the process was the same - visually follow program flow, and analyze it for errors.

@david_chisnall @futurebird when I was a teenager I developed my first Doom source port and it was how I truly learned to program in C. Having an existing codebase of good code to work within is a godsend because you are constantly being subconsciously taught what "good code" looks like. Plus yes, everything you've said here too: most of the work involved in programming is about changing code and not just writing it. Learning to debug, read code and reason about it are all essential

@mhoye @futurebird @david_chisnall It is always the same: Six months from now, the models will obsolete the humans they're hiring now.

I don't know* why I keep freaking out and getting terrified and depressed now, 36 months into "programmers will be gone in 6 months".

*) I suppose I do know; it's because I have an anxiety disorder.

@futurebird do your students think that writing code is like writing a word document and you are the “auto-correct” typo utility? ‍️

@futurebird @david_chisnall
Yes! AI-LLM is a “word calculator for grammatically correct responses”. No intelligence present

Intelligence = ability to acquire and apply knowledge and skills.

@futurebird I feel like one of the things that annoys me about LLM use is that it’s not being treated in that same manner. You need to be able to check the output of your tools. To use a calculator, you need to understand what you’re doing with it and also the underlying domain: that’s math and whatever you’re trying to apply math to.

There’s other things like the nondeterminism and waste and ethics and how well-formed language biases the brain to thinking the thing that produced that language is both conscious and has thoughts.
@david_chisnall

When I was in high school, the introductory programming class didn't let us use a computer for about a month. Instead, we would get programs on paper, and would be asked to trace out what they did.

When I taught friends how to code, I was less strict, but writing out the steps on paper was still a big part of it. I also wouldn't let people make changes to their code without first telling me what they thought was broken, and predicting/explaining what the change would do.

I think it's a very effective approach.

@futurebird @david_chisnall I use math that I wasn't taught at all in school like matrix multiplication all the time, but I have never once in my adult life used the two digit times table I was forced to memorize. I can just type any multiplication I want into a random box on the computer I sit in front of all day and odds are it'll give me the answer.

The non-memorization-based stuff I did learn in school like trig does come in handy but all the memorization is entirely useless.

I did take a few classes in high school that were called "computer science" but were actually just very basic introductions to Java programming. (After I had already been self-taught programming for a few years at that point.) They had the calculus teacher who was smart but didn't really know how to program teaching them, and I distinctly remember him and the kid who sat next to me not being able to figure out why that kid's code wouldn't compile and it was something like

if (x == 1) && (y == 2)

Anyway the reason that people need to learn stuff that AI "can do" is that "AI" can't actually do those things. It can make stuff that's in the shape of whatever, but it's not actually comprehending the text it's writing or knowing what the apple it's drawing a still life of is or intentionally writing the code that you asked it to write.

It's just copying stuff the company that made it stole from the internet. If nobody was publicly making art or writing or programming, the company would have nothing to steal and plagiarize. (Although that's definitely not a reason why people should make art.)

Generative AI is a very impressively space efficient lossy compression algorithm for a database of numbers and nothing more.

@futurebird @david_chisnall
Except only Marketing believes that LLM / Generative AI can write code or write essays.

It's either plausible junk or plagiarism.

@futurebird I used to teach machine learning to Masters students — Masters students who did not have a coding background. So we were basically starting from square 1. The were older so they didn’t interrupt, instead they would sit there with non-functional code and do nothing until I came around and asked how they were getting on. I have a puzzle solving/tinkering nature, so it was eye opening to me to see that a lot of people don’t share that at all; and must learn it as a skill instead.

@futurebird
CWSL: "Compiler Warnings as a Second Language"

@futurebird Wow! A touchy bunch.

You already have a ton of replies here AND know what you’re doing so you don’t need me (random internet person but former professor with lots of experience teaching anxious undergrads coding) to chime in.

FWIW when teaching I tried to normalize “errors” even by referring to them as “messages” because calling it an error feels like “oh I already did something wrong”. But “errors” only happen because computers are truly not very smart; an error is when it can’t understand our instructions, what we are telling it to do. The human can’t make an “error”; only the computer does. Anyway I’d also shorten the cycle between coding and compiling - having students re-compile/run the program after every tiny little change shortens the feedback loop, normalizes “errors” (like exposure therapy), and makes troubleshooting easier.

Again you don’t need my input at all — thanks for the flashback to my teaching days!

Best of luck with this group! ️

@Bumblefish

Exactly. To the untrained eye, imperfect maths can look like good-enough maths. A compiler error is unambiguously not a good-enough anything.

@futurebird

@jenesuispasgoth Nailed it: "copying without thinking" . @futurebird

@futurebird@sauropods.win I don't know your class or students, and all of this might be stuff you already do or not relevant at all, but here are some things that popped into my mind from my own experiences, in case it's of any use:

• Recent CS education research suggests that functions are one of the hardest intro programming concepts for most students, and shouldn't be introduced till some other concepts are mastered first (search "concept inventories for introductory computer programming"). It can help to first practice associative arrays/hashtables/dictionaries. Modularization into functions is harder still. The fact that many of your students are struggling with creating a function might indicate that they haven't mastered the concept yet and you might do well to back up a bit. If they need some remedial work, I cannot praise Parson's puzzles enough
• When running a lab-style section, where students are actively working on something with your support, I think it helps to interleave lecture time, work time, and debrief time. When you lecture, lecture rules (including controlling when interruptions can happen) apply. Work time is when you let interruptions happen more freely as you walk around to see how folks are doing. When I run such things I tell the students at the beginning of the section what the plan is. After a week or two they get it. I think it's useful to keep each work session on the shorter side, 10-15 minutes, with a well-scoped task and well-defined goal, and then have a debrief afterward where students can describe their experience, vent, brag, what have you. That way they know they'll have opportunities to talk and might be less inclined to shout out randomly
• If you don't have assistants to help you, recruiting other students to help field questions can be very effective. In the past I've had success dividing students into pods of 2 or 3, but only after observing the class for a few weeks. I strategically designed each pod to have at least one student who seemed to be on top of the material and another who seemed to be struggling. This setup requires communicating with the students regularly and adjusting the group assignments throughout the course, but it can lighten the load quite a bit, especially after the students get to know each other. I design classes such that the first few weeks are for setting the stage and warming up, and for me to get to know the students
• I've found it can be helpful to tell students some variation of "I know it's frustrating that your code doesn't work. Even today, code I write doesn't usually work the way I want on the first go. This is an experience you're likely to have the rest of your life when writing code. One thing to take away from this course is how not to be set back by this feeling. It's a normal part of the experience of coding, and it's telling us something". If that lands you can follow up by asking them what they think their frustration/struggle/what have you is telling them. The self reflection can be helpful and you can learn important things about your students this way (it can also lead to awesome discussions). Some students react very positively to hearing that this is a normal part of the process (they think there's something wrong with them, or that they are doing something wrong, if they're feeling frustrated).

Hopefully something in there is of use. I have references for concept inventories and Parson's puzzles if those would help. Good luck!