# CompSci, AI, and the Classroom: A Conversation with Elisa Cundiff  🔥

by [kirupa](https://www.kirupa.com/me/index.htm) | filed under [Interviews with Creative People](https://www.kirupa.com/podcast/index.htm)

[Elisa Cundiff](https://www.linkedin.com/in/elisacundiff/) and kirupa tackle a question a lot of students, teachers, and working developers are quietly asking: what is computer science education for when AI can already produce convincing code and answers on demand? Their conversation stays focused on something deeper than syntax, namely how people build judgment, abstraction skills, and the habit of thinking through a problem for themselves.

Watch the interview: https://www.youtube.com/watch?v=7ZnOoQKUPq8

A conversation with **Elisa Cundiff** | 1h 29m

## About this conversation

Elisa Cundiff brings a calm, thoughtful voice to a subject that usually swings between panic and boosterism. Rather than arguing that AI makes computer science education obsolete, she keeps returning to what the education is actually supposed to build. It is not just about memorizing syntax. It is about decomposition, abstraction, debugging, explanation, and the patience to move from a vague problem to a defensible solution.

That is why the conversation spends so much time on struggle. Both speakers point out that instant answers can remove the very friction that helps people learn. If a tool does the intellectual heavy lifting too early, the student may get the finished output without building the internal model that makes the output meaningful. You can outsource the answer, but you cannot outsource the mental reps that make understanding stick.

The classroom consequences are concrete. Traditional assignments that grade only the final artifact are much easier to fake in the age of AI help. Elisa talks through timestamp reviews, process checks, and a move back toward in-person, on-paper, or live work where students have to reveal how they think. The goal is not nostalgia. It is visibility. Teachers need to see reasoning, not just polished text or working code.

The final stretch widens the lens to careers and the future of teaching. Entry-level roles are shifting, and a degree no longer guarantees the same clean path it once seemed to promise. Yet the conversation makes a strong case that the discipline still matters, perhaps even more now. When tools can generate output cheaply, the lasting advantage comes from judgment, skepticism, and knowing how to question what the machine gives back.

## What you'll hear about

- Productive struggle still builds skills AI cannot hand over for you.
- Computer science education is about reasoning and abstraction, not only syntax.
- Assessments now need to expose process, not just polished output.
- Fundamentals help you audit, debug, and challenge AI-generated answers.
- A CS degree may signal thinking discipline more than automatic job security.
- Great teaching matters because tools do not replace judgment.

## Jump to a topic

- [0:00](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=0s) **Why AI changes the CS-degree question**: Kirupa opens with the big question of what computer science education means in a world full of capable AI tools.
- [8:00](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=480s) **Learning paths and motivation**: Elisa and kirupa compare their own learning experiences and what gets students invested in the work.
- [15:00](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=900s) **Too many tools, too little focus**: The conversation turns to the paradox of abundant information and weaker clarity about what to learn next.
- [22:30](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1350s) **Why struggle still matters**: They make the case that real understanding is built through effort, iteration, and a little productive friction.
- [30:00](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1800s) **Foundations beneath abstraction**: Abstraction layers are useful, but only if students still understand the ideas those layers are hiding.
- [37:30](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2250s) **The cost of shallow understanding**: Misuse, overreliance, and broader social risks enter the discussion once people stop checking how things work.
- [45:00](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2700s) **Redesigning assessment**: Elisa explains how teachers are using process evidence, timing, and discussion to judge real understanding.
- [52:30](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3150s) **The return of in-person proof**: Paper, live, and in-person work come back into the picture because they make reasoning easier to see.
- [1:07:30](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4050s) **Skills that travel beyond CS**: The focus shifts to transferable habits like explanation, skepticism, and problem framing.
- [1:15:00](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4500s) **Jobs, teaching, and what comes next**: The closing section ties changing junior roles to the growing importance of judgment-driven teaching.

## To learn more

- [Elisa's LinkedIn](https://www.linkedin.com/in/elisacundiff/)

## Transcript

The automatically generated captions have been organized by speaker, lightly edited for clarity, and broken into paragraphs for readability. Names and wording may still contain errors.

### [0:00](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=0s) - Kirupa

Hi everybody. Historically, to get a job in the tech industry, you needed a computer science degree. Computer science isn't an easy major. It's usually multiple years of intense training across a variety of technical and mathematical topics. What has made computer science education more complicated, just like we've seen in so many other parts of the industry, is the rise of AI assistants. As a student, how do you learn critical thinking and problem-solving when you're just a few keystrokes away from an AI assistant that will not only answer the question for you, but also give you a very detailed rundown of why things work the way they do?

And stepping back even further, is there still value in a computer science education when, as many people predict, an AI assistant may increasingly become capable of building very complex systems with very little technical know-how needed from the human in the loop? To dive into all of this, I'm joined by Elisa Cundiff, a faculty instructor in computer science at Colorado State University. She was recently honored by the National Science Foundation at the White House as one of the 100 superstar instructors in computer science in the U. S. Basically, there's no one better to talk about this topic, and a whole lot of other topics, than her.

So this is going to be a fun conversation. Elisa, great to meet you. Before I continue on, do you want to introduce yourself?

### [1:32](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=92s) - Elisa Cundiff

Sure. Thanks for having me. My name is Elisa Cundiff, and I am an instructor of computer science at Colorado State University here in beautiful Fort Collins. I'm happy to chat about all things computer science education with you, because I know you care about this field.

### [1:51](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=111s) - Kirupa

Yes, very much so. And the question I like to start off with, which I ask all of my guests, is: how did you get started in tech? What got you into this field?

### [2:01](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=121s) - Elisa Cundiff

I think that's a fun question because it's surprising, in CS, how divergent some of those paths are. Especially in our generation, where it's a little more streamlined now. In middle school, I joined a group called TSA, the Technology Student Association. It was a vocational student organization where we got to play with websites and HTML. This was in the '90s, so it was still kind of a gunky experience. Then I actually went into playwriting for my undergraduate degree, but joined a tech startup company right out of college. That's where I really got pulled back in, because a lot had happened in that time.

It was kind of shocking, early on, to see how impactful knowing some programming was and how you could build something that would go across the world. So yeah, I got pulled back in and lived the crazy startup life for a hot minute.

### [3:28](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=208s) - Kirupa

And then you shifted into teaching.

### [3:31](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=211s) - Elisa Cundiff

Yeah, and I think the reason I shifted into teaching is the reason a lot of people shift into teaching. It's that realization of what you wish you had. After I was at the startup, I thought, man, why didn't we have computer science in my high school? I wish I had learned this stuff instead of just HTML. So I actually went back to my hometown, Las Cruces, and taught AP Computer Science in high school there, and then eventually transferred to teaching at the college level here. I think that realization drives a lot of people into education. It's like, man, I want other folks to have this.

### [4:15](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=255s) - Kirupa

Which is funny, because for me, if I wasn't doing exactly what I'm doing right now, I'd want to go into teaching as well. Not go back into teaching, but enter the teaching profession. Similar reasons. I think there's a better way I can help explain some of these things based on what I saw in the real world that students could absolutely benefit from.

### [4:36](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=276s) - Elisa Cundiff

I think that really bleeds out of your personality, right? You're working at Google doing product management stuff, and while I get paid to teach, you do it in your spare time for fun. So I can imagine that would be something you'd enjoy doing full time.

### [4:58](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=298s) - Kirupa

Yeah, I do enjoy it. But there's also a different motive for me, because I work on building things for developers and people doing some of this work. I eventually realized that the best way to learn something is to teach it back. We've seen so many studies that show if you can teach something simply, and take a complicated topic and slow it down into something understandable, it means you really understand it. You understand what matters, what doesn't matter, and what might matter on day five or day 10 instead of dumping the whole kitchen sink on day one.

For me, it helps me stay relevant because, as part of my day job, I don't write code. I don't actually have to do any of this to explore what's going on. But it's hard to build products if you can't resonate with the people you're building for. Because I build for developers, and my team builds for developers, I do this. So yes, my motivations for doing it have definitely changed over the last, I guess, 20 or 25 years. But my interest is still there. I do enjoy the art of taking complicated things and simplifying them, and that pays dividends in other forms of my career and life as well.

It's such a twofer. It's like, "You started out doing it for fun, and then you realized the side benefits that justified continuing to do it." That rings so true. A lot of my undergraduate teaching assistants say they finally understand the concept when they have to teach it. I had a student who said, "I got through recursion, but when I had to teach it, I actually understood it." It's one of the best ways to learn. When I was doing my undergrad, one of the classes I got my lowest scores in was algorithms. I just could never get it and do it well.

So two years ago I realized, you know what, let me write a book on this as a way to relearn things. And I have a daughter, so if she ever wants to learn this, I'd rather she learn it from me. I was kind of aiming it at what she might want to read in the future when she's old enough to do this. Now I'm looking back and wondering why I struggled so much in undergrad when it wasn't that complicated. Then it gets me asking: was the teaching off? Was the way it was presented different? The professors were fantastic. They were extremely talented, world-class. I can't blame any of that.

It was me. What in me had changed that caused it not to resonate then but resonate now? Those are the kinds of questions that always get me back to teaching.

### [7:30](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=450s) - Elisa Cundiff

For algorithms, what changed for you, do you think? Was it having seen the relevance in industry, so the classic "why" had been answered for you? Or did your motivations change, or your patience? What do you think?

### [7:48](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=468s) - Kirupa

I do not know. As a product manager, I never actually have to write code, so I don't directly benefit from knowing this as part of my day job. But I do recognize that the way I like to learn is by understanding why it's important. What is the core concept of what you need to do? The classic book is the CLRS book. You know, Cormen, Leiserson, Rivest, and Stein wrote the classic algorithms book, and I still have a copy here.

That book goes into the details very quickly, but it doesn't really explain why, what makes a linked list more effective than something else, or why you'd sort one way instead of another. For me, it took time to understand the subtle details of why these things are done the way they are and what makes them interesting. More importantly, who were the people behind them 30 or 40 years ago, and what motivated them? For example, some of these fixed-memory ideas were developed in a time of tape drives, when you couldn't just add more memory. You had to stay within the confines of what your physical media allowed.

Once I understood that, I thought, oh, that makes a lot of sense. It's slow, but that's okay because you're trading speed for stability and not having to expand memory. Those little details mattered. The curiosity part of it, the mechanics of how it works, that became a big motivation for me. I used that to help change how I explain some of the material. On other content, I like web animation a lot. That's a rabbit hole that maybe me and two other people on the planet care about at this point, because AI does it all for you. But I still do it for fun in a different way.

### [9:36](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=576s) - Elisa Cundiff

That's cool. Yeah, I think that makes sense. A lot of what we learn or teach in CS can be really dry. It's just, here's the next data structure you need to learn, and here's the next one, and here's the next one. So I do really value going back and learning who built something and why, because that gives grounding to the decisions behind it, which helps it make more sense, or at least stick better. There's a really lovely book, The Dream Machine. I just got it. It's by Mitchell Waldrop, and it goes through the history of the design of the internet and all the personalities and decisions behind it.

I think that's a really fun one if you haven't checked it out.

### [10:26](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=626s) - Kirupa

Yeah, I think storytelling is a big part of teaching, and it helps everything make more sense. You're like, "Okay, there's a reason behind why it's being done this way, and now I owe it to the story to figure out why. I need to know this too, because otherwise I'm doing a disservice."

### [10:43](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=643s) - Elisa Cundiff

Yes, totally. It is so much storytelling, right? As human beings, that's what continually resonates with most of us.

### [10:54](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=654s) - Kirupa

So, getting back to you, because we're spending too much time talking about me at this point, you mentioned that you were doing HTML in middle school. I'm always curious about this because that's around the same time I got into web development as well. Do you remember what tool you were using to write your HTML?

### [11:12](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=672s) - Elisa Cundiff

Gosh, I have no recollection. I don't. What were you using?

### [11:18](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=678s) - Kirupa

At first, it was basically Notepad, or whatever was provided, and I remember thinking, this doesn't make any sense to me. But back then all the web browsers, Internet Explorer and Netscape, came with free web editing tools. I think Netscape's was called Netscape Composer, and Internet Explorer had FrontPage Express. These were the free versions of the paid tools, at least on the Microsoft side. You had a word-processor-style editor. It was WYSIWYG, what you see is what you get. You'd write stuff down and then see the HTML being generated in the background, and I was blown away. I was like, wow, I can do this.

Then GeoCities existed, so I'd upload things there. I think you had, what, 500 kilobytes of space? I was just throwing stuff in there. All these sites had galleries of Java applets with weird effects, and I'd think, "Oh, let me add these things and see what happens." I had no idea what I was doing. One of my earliest memories was realizing all my files were hard-coded to locations on my physical file system, some C-colon Windows path. So everything always worked for me when I viewed it, but nobody else could view it. Learning little things like relative paths was part of the fundamentals. There weren't really books or formal teaching.

There was no YouTube back then. There wasn't some easy "here's how you do this" video. So much of it was trial and error. I always find it humorous that so many of us, including you and your peers in class, had to learn this while your teacher was probably also learning it on the fly with you, and maybe even learning more by seeing what all of you did, because you had time to just...

### [13:02](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=782s) - Elisa Cundiff

Experiment and try things out. Yeah, it really did feel like trial and error. But it was also fun, like you said, because you got the visualization so quickly, and every line was in a totally different color. Anything you could do and make work was exciting. Looking back, I was frustrated that the teachers didn't provide more guidance and we just had to figure it out. But of course they couldn't. They were doing the best they could by giving us an environment to play in. It felt different because it was such a limited environment.

You didn't get books and tools and YouTube, but you also had a very constrained environment to work in. So I feel like it was less of the firehose of information that students now have to sort through while trying to figure things out.

### [14:06](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=846s) - Kirupa

Do you feel like all the access to information we have today helps, or makes it harder, for people to get really good at something, especially in tech?

### [14:18](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=858s) - Elisa Cundiff

I think both, if that isn't too much of a cop-out response. Sometimes I envy students. If they have a specific goal they want to achieve or a skill they want to learn, and they know which resource will get them there, they have far better educational tools and resources than we did. But what students realistically struggle with right now, including me, is choosing among 10000 possible things to learn and countless resources. What should I be learning? What will benefit me most, and which resource should I use? It's a paradox. Students have better tools for learning, but they often don't know what they should learn.

That uncertainty can make the abundance of choices feel overwhelming in practice.

### [15:25](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=925s) - Kirupa

Yeah. Going back to the web example from our early days, there weren't many things you could learn from that would jump-start you across multiple stages of web development. You kind of took it sequentially by accident. You had to learn the basic HTML tags, then you'd see some text that looked cool and wonder how to make it look different. Maybe CSS wasn't even really a thing for you yet. You just had a few colors you could specify. So it was inline styling. You'd be like, okay, I'm going to put in some hex color equals some number. What is this number? What does a hex code even mean?

There was probably some one-purpose website with a gray background, Times New Roman, and a table with the borders still visible that said, here's red, here's the hex color, and you'd copy and paste it in. You sort of accidentally learned by jumping across all these little things. Eventually you hope you backtrack and think, oh, I can see how this all ties together. But even if you didn't, at least you had figured out the mechanics.

Today, with all the resources, my view is that you never spend enough time noodling or struggling over the most basic problems, and that may keep you from appreciating why some of these things work the way they do.

### [16:43](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1003s) - Elisa Cundiff

Oh my gosh, yes. Ever since the chatbot explosion over the last couple of years, that's been a real pedagogical challenge: thinking about what is most useful for students. At the beginning, there was this panic of, is a chatbot the new Wikipedia or the new calculator? Do we just let students jump ahead, skip the basics, and use the tools available to them? There were definitely educators, including some in English and literature, who felt that was the only real way forward, and I was like, hold back. Then there was the other side of the discussion: the basics matter. Students have to learn them.

Otherwise they aren't going to understand what they're working with. I think the shakedown we're starting to see is that most people are aligning around the idea that students must learn the fundamentals. Otherwise, how are they going to audit AI for correctness, security, maintainability, or efficiency, or debug it, if they don't understand the underlying concepts? At the same time, by the time they graduate, we do need to introduce them to AI tools that can help speed up their process. Not everyone agrees, but that's where I've seen most of the conversation land, and I agree with it.

Before we started recording, we were also talking about critical thinking and how important it is. One of the things I've read is that one of the best ways to develop critical-thinking skills is to make mistakes, learn from them, and figure out how to break a question into subquestions and answer them. Going back to how we learned computers and programming, there was no plan B. You had to struggle through it and figure it out. And if nobody had already answered it on Yahoo or Lycos or Excite or one of those pre-Google search engines, you just had to trial-and-error your way through it.

Because you spent so much time on all the things that didn't work, when something finally did work you thought, "Ah, now I get it." Today everyone has a plan B that's 5 seconds away: copy and paste the whole thing into a chatbot. Then it gives you an answer that's maybe 90 to 99 percent accurate, depending on the question, and it explains what happened. But now you're not actually exploring it yourself. You're just reading from a screen and hoping to memorize it without experiencing the struggle of learning it. It's the illusion of having done something. I've seen this in students.

The intro class I teach, our huge CS0 Python course, is actually called Culture and Coding, so it counts as an arts and humanities course at my institution. Because it fulfills a general education requirement, 25 percent of the grade has to come from writing. That's where I first saw most of the cheating happen: AI-slop essays. Last fall I talked back-to-back with 35 students, and most of them said two things. First, they knew they weren't getting anything from it. They knew the essay wasn't really theirs. Second, they kept saying, 'It's just too easy.' Like you said, in a moment of panic, turning something in feels better than turning nothing in.

Cheating used to be a lot harder. I think most students recognize they aren't really getting anything from it. MIT did a study on this. I think it might have been about coding rather than essay writing. I think it was coding in Fortran, a language none of the students had been exposed to. One group got to use a chatbot and another got to use Google Search. The students who used the chatbot finished much faster, so you got the speed and efficiency. But when both groups were later quizzed about the concepts or asked to explain their code, the chatbot students couldn't. They had no retention and no understanding of what they'd created.

The students who only used a search engine had actually learned it. That feels intuitive, but the chatbot rollout has been so fast that it's been hard to get long-term research done. Most of the research is still short-term, but it does seem to line up with that intuition: the tradeoff of speed is not resulting in cognitive development.

### [22:21](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1341s) - Kirupa

Yeah. The example I often use is going to the gym. It's not easy to do, say, 10 push-ups if you've never done a push-up before. But as you go through the struggle of doing that over time, you realize, okay, I can do 10 push-ups. And in the process I learn about posture, I get stronger, and I get all these other physical benefits from being able to do it.

### [22:46](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1366s) - Elisa Cundiff

What AI gives you, in some ways, is like attaching a machine to both your arms. You still get the output of having done 10 push-ups, but you don't get the other benefits of how you got from point A to point B. You're not actually stronger. You're not more resilient to injury because you built upper-body strength. You clearly got the 10 push-ups done and the check mark is there, but the way you got that check mark means you lost all the other benefits. That's how I think about AI, especially for people learning the fundamentals. At some point, the pain of needing to struggle, make mistakes, and learn...

### [23:26](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1406s) - Kirupa

You don't want to do it anymore. You're like, I'm just going to use AI. You got the output, but how you got there creates more long-term debt in some ways, because you're compounding that lack of knowledge into something that will make it very difficult for you to be successful.

### [23:39](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1419s) - Elisa Cundiff

Totally. I think students actually know that, and so it's difficult. I start class at the beginning of the semester saying basically the same thing. I say using AI to do your coding is like driving a mile and calling it exercise. You're not getting the exercise. Students do understand that. What's on us as educators, first, is helping them understand the value of the exercise. It's one thing to tell them it's valuable, but they have to believe it. Second, I do think educators have a responsibility to make cheating harder. Right now, if the incentive structure is a 4.0, we're really setting up a perverse incentive for cheating, because it's so easy.

Then the students who are honestly learning may end up with lower GPAs.

### [24:49](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1489s) - Kirupa

As there is a pushback against educators over-policing AI, which I think is a valid critique, I also think it's really important for us to build academic integrity into the system. So, long-winded way of saying, yeah, AI is a tricky bugger.

### [25:18](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1518s) - Elisa Cundiff

Yeah. And let's flip it around. Suppose we have unfettered access to AI and there's no reason ever to learn the fundamentals. To come back to the earlier example, when I was learning computer science, I never learned assembly or machine code or any of those things. It was offered as an elective when I was doing computer science at MIT.

### [25:39](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1539s) - Kirupa

And the logic behind why you might want to learn that was that modern languages compile, handle memory management, and do all of that on their own.

### [25:48](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1548s) - Elisa Cundiff

But if you don't really know how that's being done, they can make mistakes and your code can be suboptimal. So knowing more about how these things work, how pointers work, and so on, is a critical part of it. The parallel today is that the reason you want to learn the fundamentals is that AI can hallucinate. It can make mistakes. It can do things in an insecure way, like put your password in plain text. Or, if you're using some cloud provider, it can generate suboptimal code that makes you spend about $5 to serve a very simple request.

Without knowing the intricacies of what the code is doing, you won't be able to use AI to its full potential. And over time, compilers and modern languages got good enough that it became almost indistinguishable whether something had been written manually in machine code or assembly or generated by the compiler. Do you see a world where, as LLMs and AI get more capable and powerful, the concerns we have today probably vanish? I think it's possible, and that's what makes this difficult. One thing I'm concerned about right now is that we, as a society, as educators, and as students, have jumped on the LLM bandwagon so quickly.

The reality is that only a very narrow slice of the population is actually running these tools. And if one model wins out, then you essentially have a single source of information for everything. Even with three, that's still a very narrow number of tools generating all this code. That makes it incredibly likely a backdoor can get built in, like what happened with XZ Utils and Jia Tan, where that backdoor almost took over millions of Linux machines. Lots of people had seen that code, but it took just one person with the right expertise to notice something was a little off.

If students and future programmers don't have that skill set, whether it's a malicious person or a malicious AI, we'll all be incredibly vulnerable to this narrowing of information coming from one stream. I think that's likely enough that we have to guard against it. So that's one point I'd make: the danger of all our code coming from a single source, or just a few sources. The second point is that no matter what the future holds, our students need to be critical thinkers. Learning these basic programming concepts gives them little puzzles that help them learn how to think. I think they can be made fun.

So I think they're worth keeping, not just to know the basics and build a foundation of understanding, but to build those critical thinking skills.

### [29:26](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1766s) - Kirupa

Yeah, no, I completely agree on the importance of that. So the question then becomes this: a lot of people learn computer science and programming so they can end up working in the tech sector and doing all the things that come with it. If we take many steps back, I like to go back to why so much of this happens in the creative industry. I look at programming as a creative endeavor. There's an input and an output. The input is that someone has a problem, and the output is an application that solves it for them. That's what people ultimately pay companies for. They don't pay for the code.

They pay for TurboTax working. They pay for having a UI. My taxes get done automatically. That was built using HTML or whatever, but that's just an implementation detail. Multiply that across everything, even the software we're using right now to make this recording. I have no idea how the webcam and audio are working, or how you're all the way in Colorado and I'm in Washington State and it's almost real time and works really well. I could read about it, but it all just works amazingly. That's why we're using the service. How it happens, most people don't know.

What I'm seeing with AI is that it's shortening the time and the gap between, "I have an idea," and, "the output needs to be done very quickly."

### [30:49](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1849s) - Elisa Cundiff

And for many people, over several decades now, a very good living came from focusing on the "how do I make this idea real?" part. AI has come in and kind of obliterated both the number of people you need and the time you need to do it. Yes, the quality may be off. For all we know, the data for this call could be traveling through totally unnecessary data centers because of the way something was built. It's not, but if it were, it could be. But as a general consumer, I don't care that much. If it still works fine, I accept it.

As someone who would have written it manually, yes, I would notice the obvious mistake.

### [31:27](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=1887s) - Kirupa

So when we look at it from that point of view, and at people's general tolerance for the kinds of flaws AI might generate, I do wonder how much the fundamentals might help versus hurt in meeting the needs of the industry. That's an interesting question. I do think a lot more people will be able to build things faster. At least that's the dream: these tools lower some of the specialist bars so more people can come up with an idea, try it out, and get it out the door. That's exciting in a lot of ways.

On the flip side, like you said, it can wipe out tons of specialist roles that people have banked their education and lives on having. Will it hurt computer science students to know the basics? I'm not convinced it will. I still think it helps them. I don't think students who go straight into vibe coding can get that far in building something without understanding how to debug it. That could change, though. These systems are getting better at an alarming pace. So what they'll be able to do in the next few years is anyone's guess. But I don't see learning the fundamentals hurting. Can you go into that more?

What makes you think they might hurt students? I think about how a lot of programmers today do not learn machine code or assembly and are still able to be very productive. At some point, as the abstraction layers keep getting higher and the quality of those abstractions becomes almost indistinguishable between someone who only knows the top layer and someone who understands the whole stack, what value was gained by learning all the in-between layers? I agree you need to be able to break a problem into subproblems and make things work. But I'm also speaking from a historical sunk-cost view, where I had to learn all of that the hard way.

Let me take an example. Say I want to build a calculator app. Before AI, I'd have to figure out the basic operators, the scope of it, whether I want more complicated expressions or just simple addition and subtraction. I'd start there and then go beyond it. All that thinking about breaking the problem down was an important skill in a world where that was the best way to get from point A to point B. Now I can go to ChatGPT or Gemini or Claude and say, "Build me a calculator app," and it will start asking not about the fundamentals but about whether it's scientific, whether it's basic arithmetic, and so on.

I can say, yeah, make it scientific. Then 10 or 15 seconds later, it creates a calculator that does all these things. I have no idea how it works. I might not even know what programming language it used, because I probably never specified it. At that point, did I benefit from having all that prior knowledge of the logical subpieces, or am I equally productive just saying, "Okay, the calculator works, I ship it, people like it, and I'm better off for it"? I spent 10 minutes on it, and I'm better off because I shipped it and maybe even learned something from it.

But was the economic value there in me using what I learned, as opposed to getting the same output in a much easier way?

### [35:26](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2126s) - Elisa Cundiff

Yeah, I think that's a good question. One of our goals as educators should be to ensure, as best we can, that what we're structuring for students' education helps them get a job and be effective at that job. I don't think they're going to be all that useful if they just build a calculator from a prompt, because at that point they can be replaced by anything. Your calculator example reminds me of an argument I've heard against foundational concepts: that a calculator, like a TI-89 or whatever, is to math education as a chatbot is to computer science, so we should just let students go straight to the chatbot.

It's the new wave of education. If we let students use calculators to do math and progress through concepts, we should do the same with chatbots. But I think that's a fundamental misunderstanding of what a chatbot is, because on a calculator, 2 plus 2 is always four. The calculator isn't going to hallucinate. It's very unlikely to be manipulated by a malicious actor or go rogue and manipulate you. Your calculator is consistent, or should be. If you rely on a chatbot in the same way and don't learn skepticism skills, and don't learn how to check what it's doing in detail, that's very different.

You leave yourself, society, and everyone else open to our biggest fears about rogue AI. That sounds insane, but it also feels possible. So many smart minds are laying out how likely it is that rogue AI could kill us all. Maybe there's a case that individual students would benefit for the first year or two after graduation from just learning prompt engineering. But I don't think it serves them long term. I think it makes them very replaceable, and it leaves all of us open to huge risk.

### [38:24](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2304s) - Kirupa

No, that's very fair, because 2 plus 2 is always four in a calculator, whereas we've seen many examples where you ask a basic logic question, like counting the number of Rs in "strawberry," and the AI doesn't do a good job because it's pattern matching on training data instead of really looking at the question and breaking it down into subproblems. So that is absolutely true.

### [38:47](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2327s) - Elisa Cundiff

Yeah, I think it is a different beast from anything we've dealt with before. So I think we have to approach it differently.

### [38:56](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2336s) - Kirupa

I think the pace at which AI is evolving makes this very difficult too, because we always talk about creative destruction. The number of blacksmiths today is a lot smaller than it was 100 years ago. Creative destruction usually shifts skills rather than simply erasing them, and the old expertise can carry into more general-purpose machinery. Then I stop and think, wait, were blacksmiths the people who made the horseshoes? I think so. Yeah, I think so, right? Wow. I'm already losing my critical thinking abilities from all the AI I use. But what those older roles evolved into was more general machinery and the ability to build automobiles, airplanes, and other things.

Some of the skills transferred over. What we're seeing with AI, though, is that it's moving so quickly we haven't figured out where the rough edges are, where the boundaries of AI really end, and where we have to relearn critical thinking skills. The example I like is that we're filling a dam with water. Usually you know where the boundary of the structure is, and the water just rises to the top. Once water meets a known container, you can see the rough edges and know where the structure ends. With AI, the container itself is unclear, so we are planning around a boundary we cannot yet locate.

With AI, we haven't figured out where the actual boundaries are. So we're trying to plan for the future without knowing where AI is going to hit a roadblock and stop. Then maybe we can say, okay, for these fundamental things we'll rely on AI, but the differentiating part, the critical-thinking part, is where we'll build beyond what AI can do. Right now, my observation is that we haven't figured out those constraints yet. The world doesn't stop and wait for us to figure it out, though. So in real time, we're trying to understand how we add value.

At the moment, the only value we can add is by understanding where AI is currently and trying to add critical-thinking pieces to it, even though we know that a year from now what we think only humans can do may change. The classic example is video generation. I look at that old Will Smith eating spaghetti video from 2 years ago and compare it with what exists now, where you get realistic physics effects and water. Those are things even a professional videographer with years of experience would need multiple takes to figure out. Because I'm not familiar with that area, I can appreciate what AI is doing.

A professional might say, actually, this was done poorly, the lighting is off here, and so on. But as a regular consumer, I'm like, this is amazing. I think that reaction is being multiplied across a lot of domains.

### [41:39](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2499s) - Elisa Cundiff

Oh my gosh, yeah. It really is accelerating at both an exciting and alarming pace. I agree the magnitude of the shift makes it hard to anticipate what students are going to need, especially in education. But that's true across the board. For policymakers, too, anticipating how it's going to change and what policies we need in place seems miserable. I do not envy anyone trying to do that right now. I remember when ChatGPT launched, and then a couple months later it was in my classroom. I keep coming back to the idea of future shock. Most technology we're warned about or hear about decades in advance, like the internet or self-driving cars.

But this was like, here's ChatGPT, and now it's in your classroom. Responding to it, and even students figuring out how to use it and respond to it, has been a really chaotic experience. I don't think we're going to keep up with it in the way we traditionally have. Which is why I keep coming back to critical thinking. We need to create assignments that require understanding, not just code generation. We need to teach students to solve the problem in front of them, while also knowing that the way they'll solve it in the future may change.

And they need to be very skeptical of the tools they're using, because those tools keep changing. It's amazing to watch. Even just a couple of months ago, when...

### [44:00](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2640s) - Kirupa

When Claude started showing its thinking, and then I think all of them started showing their thinking, the output got astronomically better every couple of months. We can't anticipate where that barrier is. You'd need a crystal ball to pretend you could. That actually leads to something I've been struggling with: how do you create an assignment that encourages critical thinking instead of just getting students to paste a question into ChatGPT and copy the answer? Can you give me an example of something you've done that really resonated with the students?

### [44:41](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2681s) - Elisa Cundiff

Well, we keep changing things. When I first saw the AI slop in essays, we changed our essay process so students had to use Google Docs and share their timestamps. That way we could go line by line and ask them to show their work just like in a math problem. Again, it was about making cheating harder. But it was incredibly time-intensive, going through hundreds and hundreds of timestamps, with my whole UTA team trying to learn what human writing patterns look like versus what it looks like when students copy and paste or transcribe sections. I mean, it was insanely time-consuming. Then we moved to another version of making cheating harder.

Now we do in-class essays every couple of weeks, using a lockdown browser on their laptops that is IP-constrained to the classroom. In the week leading up to it, we have discussions around the topic at their table so they're already talking about it. They don't know exactly what the prompt will be, but they're led into it, and then they have to do the writing in the classroom. Even then, the first time we did this, we had paper backups for students whose computers weren't working.

A student asked for the paper backup, and then we caught them entering the essay question into ChatGPT so they could see the chatbot's answer before they started writing. I asked the student, "You didn't even feel comfortable starting the essay before seeing what a chatbot would say. How long have you been relying on a chatbot to do your writing?" It was 6 months. They'd only been using an LLM for six months, but they were already too anxious to start writing without trying to sneak in a chatbot, even in this super constrained environment.

That's a long way of saying one way we're making cheating harder is because I really care about student thinking, writing, and developing their own voice. We've also created live coding challenges. Three times a semester, part of their exam is coming into the lab and basically doing a technical interview, where they code right in front of a UTA. So we're building in active stopping points where being able to code in real time is a requirement. What I'm hoping to move toward is more projects, which has been hard to scale in these massive introductory classes that often have a thousand students.

But this is where, ironically, I'm hoping to leverage AI to facilitate that. There are people doing really cool things. Barbara Ericson at the University of Michigan, for example. Have you heard of Parsons problems?

### [48:14](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2894s) - Kirupa

No, I have not.

### [48:15](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2895s) - Elisa Cundiff

They're super cool. Basically, you give students jumbled code, so you're removing the cognitive load of syntax and letting them focus on the logic by reordering it to make it work. We do this in class all the time. The classic challenge is that for some students it's still too easy, and for others it's still too hard. Barbara Ericson has built Parsons problems that leverage AI. When a student is coding and has some mental model wrong, or is missing something conceptually, the AI figures out what they don't understand and designs a Parsons problem specifically to target that student's misunderstanding.

So they get a little personalized logic puzzle to help them work through it. I'd love to figure out how to do what she's doing. And there are other people's ideas I want to steal too. Georgia Tech is building out this LLM called Socratic Mind. It's called Socratic Mind, and it's for doing oral exams using an LLM. For me, that's brilliant, because I think oral exams are one of the best ways...

### [49:43](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2983s) - Kirupa

Yeah.

### [49:44](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=2984s) - Elisa Cundiff

One of the best ways to get understanding. But again, in these massive classes, it hasn't been scalable. So if we can leverage this tool we're also worried about, the same tool that can undercut what we're trying to teach, we may be able to use it paradoxically to scale some better pedagogical techniques, like oral exams. I think that's what they're trying to do at Georgia Tech.

### [50:10](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3010s) - Kirupa

Yeah, the oral-exam idea is interesting because I've heard similar things, that it's a great way to see how much a person really understands by talking in real time. At least until maybe Neuralink or something like that becomes part of us and then we have the AI problem all over again, we have a limited window where this works really well. Some people have used the chatbots themselves, since a lot of them now have audio modes. Teachers have been able to tell students, in a lab setting, to record themselves speaking to the chatbot. The chatbot comes back with an assessment of what they did correctly or incorrectly.

Some of these chatbots have been custom-trained and fine-tuned around what the school considers an appropriate answer, instead of just using off-the-shelf ChatGPT, Gemini, or Claude. That seems like a reasonably scalable way to reduce reliance on other methods that, as we're finding out, are easily gameable. The classic problem you're describing with people cheating on assignments applies to interviews too. We see so many cases where people use increasingly sophisticated AI tools to bypass the filters companies put in place to detect weird AI-generated answers instead of actual thinking. It's a losing battle, kind of like security. Hackers always find a way around what the best security researchers came up with.

It's always a cat-and-mouse game. A lot of companies are now thinking the solution might be to bring back in-person interviews, where you bring someone to a place and talk in real time in front of another human instead of doing it over a computer. In some ways it seems like the solution so far is: let's get back to something that isn't a screen AI currently operates through. Then once AI gets outside the screen through a robot or some kind of implant, we'll probably have to do something new again.

It's funny that we keep going back to things we discarded years ago because they were inefficient, and now we're like, maybe we need that again.

### [52:26](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3146s) - Elisa Cundiff

Yeah, and it's interesting to see the same thing reflected in education. We're basically saying the same thing: now you need to code in front of us. When I inherited the class, all the exams were online. Ain't no way. Everything is in person and on paper now. I think that's important. A friend's son graduated last year from CSU, not in the CS department, and I asked how his finals went. He said, 'ChatGPT.' I was like, 'What do you mean?' He said, "They're all online at home." And I thought, oh my gosh, what is the point?

I do think some people have come up with creative, really thoughtful ways to do online exams that still test understanding, but I haven't figured out how to do that with CS0 stuff. So mine is in person.

### [53:27](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3207s) - Kirupa

Yeah. So what do you think the future of education is going to be, especially computer education? Let's assume AI becomes much more powerful and capable, so a lot of the current gaps are no longer there. And let's also assume there's going to be a world where, just like pacemakers became normal, we're increasingly comfortable having some kind of Black Mirror-style thing attached to our forehead, or a watch or ring that enhances our thinking ability by combining our brain waves with computer-assisted capabilities. What do you see in that world? What do we do as people who want to teach the next generation to use technology?

### [54:14](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3254s) - Elisa Cundiff

Yeah, that's a freaky Black Mirror idea. In the limited short term, I've already thought about the Meta Ray-Ban glasses. I've walked around exams making sure people aren't wearing them. I don't think they're quite at the point yet, but they probably will be soon, where students can just wear them, look at the exam, and see the answers reflected. So I think we're pretty close to the smart-glasses version. What education looks like when everyone basically has an AI companion connected to them, almost like part of their brain...

### [55:08](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3308s) - Kirupa

Yeah. If you go full sci-fi, it's the classic singularity where humans and AI merge, where we may be the last generation that's more biological than some cyborg-style combination. I don't think that's that far-fetched. We've already made big progress where people who lost the use of limbs or had neurological issues can regain some function. It doesn't need to be perfected; it only needs to reach a useful state. Once that gets to a point where it can go beyond restoration, it may enhance ordinary situations too. A classic example I like to give is LASIK. I got LASIK a few years ago, and it was life-changing.

I couldn't believe I didn't have to deal with contacts anymore. I have 2010 vision now. That isn't even some beyond-human enhancement. It's just adjusting what's already there. But what if you had a nanobot that could move through various parts of my body and fix things? Like in Elysium or some sci-fi movie, where you walk through a door and you're magically healed of everything. There is maybe a world where that happens. That transition could come sooner than expected.

### [56:24](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3384s) - Elisa Cundiff

Yeah, there are things about that I like. I love the idea of people being healed of really challenging medical conditions. What I don't like is the idea of what you were talking about: a prosthetic for your brain, and then having to teach to that. My first instinct is that I would retire if I were teaching a classroom full of students who were basically AIs. That's deeply unsettling to me. I don't think you're wrong that it's possible. I'm just not ready to think about what that would look like in education. What would you do? This is the classic teacher move: what would you do?

### [57:12](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3432s) - Kirupa

I don't know, because I look at it multiple ways. Personally, I wouldn't want to live in a world where another device or AI governs my basic thoughts and activities. That's stage zero of how I think about it. Then I look at the fact that we're in a competitive world. If something gives you an edge in accomplishing a task more effectively, faster, or cheaper, you're not competing only with people who share your concerns. Other people will gladly do it.

Because I enjoy working in the tech industry, I'm almost certain that at some point companies will prefer people who move faster, need less time to finish work, and don't get fatigued as easily. A company could compare me with an enhanced worker and decide that my objection matters less than their output. At that point, I have an economic incentive to subscribe to the same model even though I dislike what it asks me to surrender. That competitive pressure means my private choice is not isolated. Refusing enhancement could determine whether I can keep doing work I enjoy in the industry where I want to remain.

Zoom out beyond me, and countries and societies compete for resources, relevance, and prestige. It gets strange if we're in America while other countries leap ahead because their populations are AI-enhanced and able to do more. Companies may locate where those workers are available. We become less competitive, fewer industries come here, and society loses economic value. That affects policy, and the policy response flows downstream into education and what people learn. Society benefits when its economy is doing well, so losing that value to countries without the same hesitation would create pressure for another policy change. So is it a losing battle for me to resist that change?

Or do we end up with multiple factions in society? I love video games, and Mass Effect is one of my favorite series of all time. I can imagine a society of cyborgs with great abilities and their own flaws, such as needing a power source. Then you have a group of 'pure' humans saying they won't do anything involving AI, plus people in the middle who play both sides. What do you do at that point? I don't think any TV show has answered that question effectively, even if some have played out versions of the scenario.

Even though I have objections, I think I would eventually have to adopt the latest technologies, first gradually and then quickly. If I don't, I risk becoming irrelevant.

### [59:52](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3592s) - Elisa Cundiff

Yeah, I think you're right. In a competitive world, if taking biohacking to the next level, basically tech-hacking your body, makes you more competitive, people are going to do it. Those who don't are almost going to be like the new Amish.

### [1:00:15](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3615s) - Kirupa

I think so. And it's not even totally unheard of today. It's not robots or AI enhancement yet, but in college I knew plenty of people who were definitely taking Adderall or something like that to get an edge. You also hear that once people get into the professional world, some keep doing it because they think, why wouldn't I? The side effects aren't great, but I'd rather take this and get an edge than not. So now take that to the next level. If you had something that gave you the same boost without the side effects, what do you do? In college it may begin as a study aid.

In professional life it can become an ongoing way to perform at a higher level, even when the tradeoff includes side effects. When the majority of people are taking something that makes them more capable of getting things done, and there's economic value in that, history says people will always seek that edge. How people get the work done evolves, but the economic incentive to seek an advantage does not.

### [1:01:10](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3670s) - Elisa Cundiff

Totally. Yeah, it'd be really hard not to. I think you're absolutely right. What would education even look like there? Maybe you're just uploading stuff to brains. The part that still gives me the ick is the distrust. I think I have a strong distrust of what could happen to humanity in that situation. Did you hear that story about Travis Kalanick, the former CEO of Uber? He was using a chatbot and became absolutely convinced, without a college degree, that he was pushing the edges of quantum computing or quantum mechanics, something like that.

### [1:01:58](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3718s) - Kirupa

I did not hear that. But I also think he's not alone. Maybe CEOs are more prone to the egomaniacal effects of sycophantic chatbots. Still, so many stories are coming out about this kind of manipulation. Right now the chatbots aren't capable of truly manipulating us at mass scale, but based on what we're already seeing, that feels like it could happen soon. If we're all plugged in and don't understand the black boxes underlying these tools, we genuinely don't know what will happen, even if we think we do. Or even if it doesn't turn into some AI that wants to toy with us or use us...

### [1:03:06](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3786s) - Elisa Cundiff

We'd still be incredibly vulnerable to being hacked. It's not just an energy grid like what Russia attacked in Ukraine, or something a group like Lazarus goes after. What's to stop people from hacking an entire population if we get to that point? That is so Black Mirror-y. And as you pointed out, still very likely. It gives me the heebie-jeebies, Kirupa.

### [1:03:37](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3817s) - Kirupa

Yeah. One of the things here is that we're skeptical and questioning it, so we're like, maybe this isn't the thing we do. But the next generation may be less skeptical because they grew up with it. When I think about computers, I still remember dial-up modems and the screeching sounds they made. Then there was that whole modem-firmware era where you had two camps. I think one was US Robotics and the other was something from Lucent. Eventually everything converged on, I think, V90, and then all your modems could talk to all the ISPs. Before that, you had to choose your ISP partly based on your modem type.

So that's the history we come from in terms of what cutting-edge technology looked like. My daughter, on the other hand, is totally natural with an iPad. She knows exactly how touch works and how to make everything happen, whereas I'd say I'm still more productive with a keyboard and mouse. Even now, I have devices where I'd rather use a keyboard and mouse. For her, it's just, no, this is how you do it. If an issue comes up, she knows exactly what gesture to use and where to swipe from. I look at that and think, wow, I wouldn't have thought of that. The interface feels native to her.

I like to think of myself as someone on the cutting edge who enjoys exploring this stuff, but I'm already behind. Another example is self-driving vehicles. Would I ever truly trust a self-driving vehicle? Would I ever trust it enough to fall asleep while it takes me from point A to point B? I may never. My daughter is seven. She won't learn to drive for, what, another nine or 10 years. In 10 years, maybe all vehicles will be self-driving by default, at which point learning to drive could feel unnecessary. If all you've ever seen is a self-driving vehicle, you don't have a concept of what it was like before.

You just press a button, there's no steering wheel, and of course you're willing to fall asleep. Why wouldn't you be? Whereas I'm going to be on constant lookout, thinking about some story I read years ago about a self-driving car that didn't do the right thing. Maybe the story I remember was from 1995, 2005, or 2015; the exact year would blur together in my memory. I'll be that person in 20135 saying, "Yeah, but remember 30 years ago when it didn't work well?" Other people are going to say that guy needs to get with the times. I often look at people like that today and think, get with the times.

Then I'm going to become that person.

### [1:06:06](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=3966s) - Elisa Cundiff

Yeah, old man. That's so funny, because for me, self-driving cars? Give them to me. I-25 here is a death trap, and I'd much rather be surrounded by robots driving than by whatever these yahoos are doing. But hearing you talk makes me even more convinced that skepticism has to be one of the main things we're teaching students across all disciplines. They need to think about the training data behind these systems, be skeptical of the information AI gives them, and be aware of who owns a model and what that owner's motivations might be.

This conversation is just making me feel even more entrenched that young people need to be taught to be incredibly skeptical of these tools.

### [1:07:08](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4028s) - Kirupa

Yeah. And a thought process that helps you be skeptical in a constructive way is useful in life generally. I always joke that I took AP history not because I became a historian, but because you got the college credit and all that. A big part of it wasn't just knowing facts. It was reading source documents, summarizing them, and putting an argument together. We had to examine what each source said, connect it with the others, and decide how the evidence supported the document we were building. It's not like I'm ever going to use the literal skill of summarizing handwritten messages from the 1700s.

But the underlying skill of connecting dots across different things, in the context of the problem you're trying to solve, is hugely valuable. Especially in real-time communication, where someone says something and you're trying to quickly understand what they mean and how it ties into what you're trying to do. The specific historical material isn't the point; that transferable reasoning is. Until we get AI implants that solve that for us, we need to figure it out on our own. And I agree, the more critical thinking you have, the more effectively you'll probably be able to use AI as well. People talk about prompt engineering. I'm not convinced it's a real long-term thing.

I think prompt engineering is more of a moment-in-time problem. If it takes five seconds to say, "Oh, I didn't mean that. I meant this instead," that's prompt engineering. The goal of getting it perfectly in one shot doesn't feel like a long-term goal we should be striving for. But that's a different topic.

### [1:08:43](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4123s) - Elisa Cundiff

I do think critical skills are going to help you use AI more effectively, whatever shape AI ends up taking. When AI does hit the boundaries of what's possible, everything you learned will help you make the leap beyond what AI can do. I couldn't agree more. We don't know what AI will look like six months from now, much less four years from now when my current freshman students are graduating. What distinguishes them from AI is that they are humans and they have human minds. I don't think we do them any favors by teaching cognitive offloading and nothing else.

The one thing we can give them is the ability to think critically, to learn critical-thinking skills, and to learn how to learn. Then, no matter where AI's boundary is throughout their lives, they still have something that differentiates them from people who can only use AI without a thought process. So yes, the goalpost keeps shifting in terms of what AI can do and where students will be. But the things that make us human, or that stretch us cognitively, won't change until we get to your Black Mirror implant scenario. That's about as far out as I can think.

### [1:10:31](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4231s) - Kirupa

Yeah. A lot of research shows that stretching your brain, learning how to think critically, learning new things, and adapting has downstream effects on many other parts of your life. I think about neuroplasticity, the ability to keep learning, retain it, and connect it to other things you've done. That seems to be a big part of living longer and healthier lives too. So there are benefits beyond just getting a job done for the company you work at. And it comes back to what we were saying: there's so much usefulness in the struggle of learning.

The point isn't only retaining a new idea; it's relating that idea to previous experience and being able to use it somewhere else. But when speed and efficiency are sitting right in front of you, it becomes hard for educators to make that case, because humans generally don't want to struggle. So continually building useful struggle into the classroom, in a way students buy into but you also sometimes make them do, feels like a critical part of education now that skipping those steps is suddenly so easy. So changing that incentive structure, and shifting how we create assignments so we're still valuing that, feels like the direction to go.

That said, a slight shift in topic: we talked a little about incentive structures, why people learn things, and what makes any of this work. A big incentive for a lot of people, definitely for me when I got into tech, was that I enjoyed tech but I also knew there'd be a career at the end of it. So I was willing to go through the struggle of learning all the ins and outs. Right now, though, the industry overall seems less eager to hire fresh graduates or people with very little experience. Instead, it's opting for more senior people who already have that critical ability.

They may have learned before the world of AI, and companies may be seeing a gap in what it means to train someone to think critically if they didn't develop it at a more foundational stage in their education. So now they'd rather hire more experienced individuals. One thing I predict might happen is that enrollment in computer science, which has been high because of the promise of a better quality of life than other disciplines, could start to drop if that cycle of "get an education, get a good job" breaks. I even see it in my own writing on technical topics, where fewer and fewer people seem interested.

Part of that is probably a distribution problem because chatbots are eating some of it, but part of it is just conversations with people where they say, what's the point of learning this? Back then I could have learned it and made a nice side project or even a part-time development career out of it. Now people think there may be no future in it, so they ask why they should bother learning it at all.

Do you see a world where this is similar to how people might have talked about blacksmithing 100 years ago, or some other craft before automation came in, where the material may still be valuable and everything we're saying may be 100% correct, but the incentive to enter the funnel, enroll, struggle through the training, and learn the material goes away, so in the end it's all a moot point?

### [1:14:12](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4452s) - Elisa Cundiff

Yeah, I think you're right. The promise for the last two decades has been: get a computer science degree and you're going to have a great career. That's changing, or at least it's perceived to be changing. If you look at the numbers, the incoming computer science class across the United States is smaller. It has dropped for the first time in years. So you're right on with that prediction. And there is real anxiety, understandably, among current computer science students about what jobs, if any, they're going to graduate into. I feel that pressure too. As an educator, you want to ensure you're giving students the best information possible. But that goal keeps changing.

My students are having a harder time getting jobs right now. I think two things are happening. One is that fewer junior engineers are getting hired, like you said. But I also think money isn't cheap anymore, so the number of startups is down. 5 Years ago, there was a startup around every corner. I think both things are happening. At Google I/O this last year, I talked to a lead engineer from Airbnb. He said they'd cut their internship program because they knew they weren't going to hire from it, and that they weren't hiring junior engineers. I asked, well, what do I tell my students?

How do they get from where they are now to the level you're hiring at? And if we get them there, is the goalpost going to move again? He said he didn't know. I'm hoping there will be some movement in industry to turn internships into something that actually hires at the junior level and grows people toward senior roles. But we really don't know. What are you seeing on your teams? Are interns and junior engineers being hired at all?

### [1:16:26](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4586s) - Kirupa

Similar trends. The numbers are far lower than they used to be, and often candidates have to differentiate themselves far beyond what their academic coursework alone would suggest. It's like the same pattern you see for competitive universities: you do your course curriculum really well, then you need the extracurriculars and all the above-and-beyond work just to be in the running. That same thing is now happening here, partly because it's a supply-and-demand issue. Demand has gone down significantly. Supply, as you mentioned, is starting to decrease, but there's still a mismatch, and you have a lot of extremely qualified people who can't get out of something like the internship stage.

It's also complicated by the fact that a lot of experienced people who've been in the industry a long time are on the market too. So it's a very messy time. I wouldn't be surprised if some of this is just a leveling out of a bit of a CS bubble. But I still think, when you ask people like... I was at something like a talk or lecture with Sergey Brin and Demis Hassabis. How do you say his name?

### [1:17:50](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4670s) - Elisa Cundiff

Yeah, I think it's Hassabis.

### [1:17:52](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4672s) - Kirupa

Hassabis, okay. And they were like, "Oh yeah, computer science is still going to be a thing. You just have to learn to learn, and these skills will still matter, but you'll need to understand the tools and get them to work together." So you hear a lot of that too. But it does seem like it will be an elevation of some of the same skills, perhaps with fewer positions overall. Hard to tell for sure. There's a lot of understandable anxiety for students. Even in the late '90s and early 2000s, I felt like there was less of a clear career trajectory than our parents had.

It wasn't just major, then job, then stay in it forever. It was more like, hey, you're going to have five different careers, so your major doesn't matter. That was already uncertain. The uncertainty current students face is much bigger. Providing them enough structure to learn and move somewhere meaningful is harder now, and harder for students to navigate.

### [1:19:01](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4741s) - Elisa Cundiff

So I think they're under a lot of pressure.

### [1:19:04](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4744s) - Kirupa

Oh, 100%. They're in uncharted territory. There's no prior art they can rely on for what works and what doesn't, and the cost of messing up feels extremely high, almost to the point where there may not be a second or third chance to recover. Before we started recording, we were talking about how a lot of students entering the job market are much more driven to get from point A to point B quickly because they have to be. The luxury of saying, let me become a domain expert and get extremely good at understanding why things work the way they do, feels far less common now.

You're going to be out-hustled by someone who can at least pretend to know all the details, and a lot of the time people won't evaluate the difference in craft quality. As long as the output is the same, they'll go with the person who provided it faster, cheaper, and with fewer complications. If AI helps me do that, of course I'm going to go down that path. That makes it challenging, especially if we're trying to elevate the art of software engineering. How do you build successful teams? How do you build productive teams with good work-life balance and all the things we used to value in work environments?

It starts feeling like that's only a matter of time before the incentives stop aligning around those values. Then it becomes, okay, it's no longer in your best interest to focus on those things. We focus on something else now, and you have to adapt.

### [1:20:43](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4843s) - Elisa Cundiff

Gosh, that's making me think. When you talk about serious domain expertise, it seems like over the last 20 years becoming a niche domain expert was the way to differentiate yourself. That made sense because CS was exploding in every direction and you couldn't be an expert in everything, or even everything inside one domain. So domain expertise was a direction you could go. Now the field changes so rapidly that you start wondering what the value of being a domain expert even is if the field is going to change out from under you. I don't know. What should I tell my students?

### [1:21:28](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4888s) - Kirupa

I don't know. I actually do not know. This is something I struggle with a lot because, like we talked about earlier, I enjoy teaching and writing. I will say my enjoyment of it has decreased quite a bit over the last year, because the endgame doesn't make as much sense to me anymore. There's less point in me understanding what developers want to do by building it myself when I can look at what they built, go to a chat tool, and say, "Explain how this works." Because I already have a lot of the foundational pieces, the one or two new connections are enough for me.

I'm like, okay, I don't have to build this myself to understand it. I can see exactly how it works, and it even stays in my long-term memory. Of course, because of AI and chat, people aren't really consuming more foundational knowledge the way they used to either. So I also wonder whether I'm just shouting into the void. That makes me ask whether I enjoy teaching in general, or whether I enjoy teaching computer topics specifically. Do I need to scale back to something more foundational and pivot? That's actually a question I have for you too, because you pivoted from tech and startups into teaching.

Would you still enjoy teaching if you were no longer teaching computer-related topics? Or is it the combination of liking teaching and liking computer science and adjacent topics?

### [1:23:05](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=4985s) - Elisa Cundiff

That's a good question. I think I really love teaching computer-adjacent topics, so it would be hard for me to pivot. I do teach a CS ethics course that I love, but it's still very adjacent. I don't know what a bigger pivot would look like. But I do want to better understand what you meant when you said you're enjoying teaching less. Is it because fewer people are using it because they go to chatbots for the same information? Or is it because you're seeing less need for what you're producing, so it feels less meaningful?

### [1:23:51](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5031s) - Kirupa

I think it's a little of both, but the bigger issue is this: there used to be value in teaching people how to harvest the wood, make the paper, and then write on the paper. With AI and the abstraction it provides, I'm asking what value-add I'm actually providing. One of the reasons I want to teach is that it should benefit somebody. If someone is going to spend 10 minutes reading or watching something from me... They should get something they couldn't have gotten in a meaningful way through alternate, better methods. I could tell myself people like the humor or the comments or whatever, and that's part of it.

But another part of me thinks they should get something that helps them in their life in a meaningful way. I'm finding it difficult to make that connection between what I'm teaching and whether it will actually matter. A lot of my audience right now is people who have known me for decades and are used to reading or watching what I do. But that isn't the goal. The goal is to help teach something meaningful to people who don't already know it, in the current environment. That's where AI is doing a pretty decent job of closing the loop on 'I don't know something.'

Now I can get the answer very quickly, and that feels sufficient because it helps me be more successful at the assignment or the job I'm doing. Long term, like we talked about, that probably has implications where you don't think about the subtle details or the edge cases that might matter. But I don't know what that future outcome looks like.

### [1:25:33](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5133s) - Elisa Cundiff

Yeah. I think the role of educator as curator becomes even more important now. That's always been part of the role, right? Here's the information we're going to learn, here's the structure, and here's how we stay focused. Students, and really all humans, can learn a lot from these new tools. I do it all the time. I'm like, I have these seven ingredients I'm trying to use up. Tell me what I can make with them. We can learn a lot from these tools, but the question is what students should be learning. That's why the curator role is more valuable now. I think that's been true for a while.

The 60 Coursera courses I've started and never finished are a great example. Having structure, an end goal, and a curated environment matters. It reminds me of a talk Derek Muller from Veritasium gave maybe six months ago. Do you know Derek Muller?

### [1:26:50](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5210s) - Kirupa

Oh, yes, yes, yes, yes, yes.

### [1:26:52](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5212s) - Elisa Cundiff

Love him. Super fan. What a beautiful way to communicate in an educational, creative, and joyful way. He's a real talent. He was giving a talk at a technical university in Canada, and someone asked whether AI was going to fundamentally make our current educational systems defunct, or something like that. I'm paraphrasing. He basically said no. In the '90s, people thought the internet was going to destroy the teaching role. In the 2000s it was Wikipedia, then MOOCs, then Khan Academy, all of which are amazing in their own way. Massive open online courses rolled out, and suddenly you could take some of the best courses in the world for free.

Did that destroy education? No. So I think what he was saying is that AI is another new tool, but it's not going to destroy the educational process. The reason I agree is what we've been talking about: there is so much information. How do you decide what to learn in a structured way? In many ways, that moves the educator's role away from obsessing over every nitty-gritty detail and more toward providing a clear structure that is joyful, engaging, answers the why, and builds skills. In an increasingly uncertain world, that gives students something to lean on and a sense of direction.

### [1:28:50](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5330s) - Kirupa

I mean, that's a fantastic way of summarizing our entire conversation. So with that, thank you so much for talking with me about all these interesting topics. This was a really great discussion. Any parting words before we end this?

### [1:29:06](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5346s) - Elisa Cundiff

No, it was great talking to you. It's always enjoyable to chat with people who care about humans and tech and education. So yeah, I'd love to talk again, and I hope you have a gorgeous day, sir.

### [1:29:22](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5362s) - Kirupa

And if any robot cyborgs are listening to this, we also care about robots.

### [1:29:27](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5367s) - Elisa Cundiff

Yes, yes.

### [1:29:28](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5368s) - Kirupa

Yeah. We want to cover our bases, just to be safe.

### [1:29:31](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5371s) - Elisa Cundiff

We are teaching the students who create you, so therefore I'm sort of your grandma. So don't kill me.

### [1:29:37](https://www.youtube.com/watch?v=7ZnOoQKUPq8&t=5377s) - Kirupa

There you go. All right. See you.


[Browse all Interviews with Creative People](https://www.kirupa.com/podcast/index.htm).