Post

Conversation

I'm teaching databases this semester at Berkeley. My students all seem unusually brilliant. Not many go to office hours, and not too many folks post on the course forum asking project questions. Weirdly, the exam had the lowest recorded average in my 10 semesters teaching it.
David Watson 🥑
Post your reply

Put another way, I'm pretty sure a lot of Computer Science students are using ChatGPT to complete their coding assignments instead of actually doing the assignments themselves. If true, it's a huge problem. The process of learning debugging is critical to growing as an engineer.
IMO the effect of this may be that courses will have to give even *less* scaffolding, so that students can't GPT the entire assignment. The design aspect of projects may also become even more critical. Theory is going to become an even more important filter for competency too.
Here's the problem with just GPT'ing your way as a student. In production-level systems, you will be dealing with things where THIS DOES NOT WORK. You *cannot* just use GPT as a crutch, partly because you won't even have the context needed to give the prompt to solve it.
For example: there will be times when the code is so gnarly and the error so insanely difficult to debug that you cannot GPT your way through it. The only way to fix it is to set a breakpoint, step into the debugger, and see what the values of certain variables are.
When setting up a codebase, there will be times when GPT fails. It's great to use GPT as an assistant — I use it all the time. You cannot use it as the *primary means of development*, because you're functionally not understanding what you're doing (other than copy paste).
Now, you can ask: "what if my tasks at work are simple enough to where GPT does solve it all, easily? Can't I just use it for that?" Congratulations. You may have discovered the path to being unemployed. If the AI does everything you can do, *why would they keep you around*?
For a CS skillset that can't just be replaced by an AI, you have to learn fundamentals. That lets you do tasks that an AI still can't, because your brain is very powerful. If you're ChatGPT'ing your way through it, you're not building any of those connections for later use.
my sincere apologies for the long, borderline-preachy rant, but if you're reading this far into the thread, you're probably already interested in the topic so I figured I'd go all the way in explaining my reasoning behind my thoughts here.
One key point, everyone: The reason AI usage is different from the boomers ranting "in MY day we used to use punch cards and wrote code in assembly" is that here, students are not just outsourcing the work. They're outsourcing critical parts of the *thinking*.
Parts of it may well be deemed "outdated". But the reason college curriculum is structured as it is instead of being a grand industry tour on the Hot Topic Of The Day is that by teaching fundamentals, you teach students *how* to think, learn, and work. AI just bypasses that.
In-class labs are impossible to put in for this semester — rooms are reserved! I do think perhaps this is something worth investigating for future semesters, if I ever come back to teach (though I don't think I'm going to have time moving forward due to a few work/family reasons)
Good thread. I find LLMs reflect a shallow understanding of database internals, FYI. For instance, just a wild idea - LLMs aren't great at fine grained concurrency. So instead of asking them to write a Bloom Filter from scratch, you could require minimal-latency row inserts,
The biggest issue we run into here is that like...I'm sure classes can make things more complex, but it feels like students are basically prompting a language model to skip past learning the basics, and these are basics they *need* to actually know. There's no real substitute.
The college days are coming to an end soon. 🚀 Fun Fact: Our graduates are making serious bank! 💸 💡 HVAC: $32/hr = $66,560/year 🔌 Electrician: $29/hr = $60,320/year 🚽 Plumber: $27/hr = $56,160/year All from our 100% online courses that take just 100-120 hours to
What does “teaching databases” mean? Roughly? I took an intro sql class once covered select, creates, sum, float, joins, triggers, stored procedures
sql, indexes, disks/files/buffers, caches, joins, query optimization, transactions, recovery, parallelism, distributed transactions, kd-trees, etc
speaking as a student taking a csen course -- I indulged in AI resources (Claude ) because whenever I actually tried to ask a person for help, such as the TA for the lab, they'd tell me "just Google it" and it turns out AI not only answers my questions, but can elaborate too
I would just make the assignment / take home stuff way harder and make it clear they can use AI. I feel that may be one avenue to take it?
See the problem with this is that someone has to design and grade it. It can get surprisingly difficult and manpower-intensive to do that when there are 400 kids in the class.
Did the pandemic hurt their social skills in asking for human help? They may feel more comfortable asking a computer instead of the knowledgeable person made available through tuition. I wonder about this sometimes as a partial issue for these kids.
All of education is at a massive inflection point. Students are no longer convinced that knowledge acquisition will help them advance their lives. The cheat codes have become too powerful.
A mystery, right? My students were the same. All doing very well on homeworks and on online quizzes but oddly tragic on paper exams when their phones are taken. A guy told me "if I need this, I'll hire someone to do it for me". I was teaching Math 1000. High school math!
I genuinely don’t understand this. Wouldn’t it make sense to only train students to problem solve at & beyond the limits? You simultaneously say “if you ai can do everything you can do, why keep you around?” and also say “you need to learn what ai does” ?? doesn’t make sense
Show more
Gonna be a bright future out there for people with actual programming skills to charge >500k/yr to maintain all the poorly written legacy the "vibe coders" are about to sell to customers.
What will actually happen is degrees will become useless as a certificate of knowledge as courses become piecemeal and adults continuously take courses throughout their career as they need to learn. Professors will be reduced because every student will have the opportunity to
Echo chamber with smart people not understanding doing the work is how you learn. Relentless practice, every great athlete and musician. What is weird is this ‘clever’ chatGPT group cannot be convinced by those who worked their own problem sets.
I hear you but I do not understand where the problem lies. Students are paying you for structured learning. You're providing it. If they're cheating, they're cheating themselves. Not you. It will simply make the ones that do the work that much more employable. It's the sacrifice
They’ve learned how to survive without engaging deeply, are probably relying on AI, past solutions, or just sheer confidence until the real test exposed the gap. Modern students are smart, but in a different way. Efficiency > Understanding.
chatgpt/grok are both amazing tools for learning, but people instead use it to do all the work for them WITHOUT having to actually understand it. that’s where the problem arises.
At the university of Illinois in the 90’s I had several classes where we simply had to write code for our finals… in pencil. No way to cheat. The only way to pass was to have already written thousands of lines of code throughout the semester.
Does the curriculum allow pop quizzes? I personally hated them, but it would show earlier the issues going on and those who really want to learn will possibly change course and learn?!? Or does this type of student no longer exist?
did you try telling your students that it's important to solve (some) exercises without external help? students may not be aware which parts they are supposed to learn and which are open-ended/as long as you get it right even with help it's fine
When the calculator was introduced Im sure the need to multiply large numbers by hand was greatly reduced for productivity purposes. High schoolers 100 years ago may be better at hand calculations than high schoolers today. This transition however does not greatly reduce the need
Lack of engagement. Lack of focus. These kids that were in high school during Covid lack something. My fiancé is a professor and is struggling with her students this semester. Many don’t show up to class, and feel like they should be given all of the material to go study on their
Most of the students are not nearly as brilliant as they seem. They are lazy and they cheat as much as possible. They do the bare minimum to skate by.
I think the "my students all use AI" issue is the equivalent of math teachers being mad that different use calculators. Although what you're saying is true now in the future AI will easily surpass human software engineers. Well college is a pointless scam at this point anyways.
This is great news for older developers who understand these topics. It’s great news because it means that the alleged new competition in the market isn’t really competition at all, and we can maintain our jobs or take the jobs that would’ve been offered to these children because
Because students see stories about other students that vibe coded an app that’s making $1m ARR. There’re no stories about rich students who know CS theory well.
DC Metropolitan area community college student hear: That’s not true for the entire generation; We’re consistently being told that we’re under performing. They are disappointed in us. We’re all working 30 hours a week. And full-time in school. We’re dead.
The problem is we are right now in the typical transition phase and in five years GPTing will outperform every programmer on the planet. In other words: you’re teaching a skill with very short useful live in 2025. Why invest 10,000 hrs to master a quickly perishable skill? To
The fundamental problem is that the exam is absolutely irrelevant to a real world situation. With AI, you need fundamental knowledge of something—you need to know what to ask for. You describe what you want to the AI at whatever level of detail you want, and the AI spits it out
You got a handful of either bored students or forced to take this course. One thing middle school and high school fail at is teaching to have passion for a subject. It’s takes a certain breed to enjoy using mySQL.
old guy here with a debugger that i use most days to find issues with production c code. LLM s are very useful to look up api. or simple toy stuff that we might overlook. basically another tool to complement your core skills . debugging via prompting can be a powerful like
I was speaking to my daughter about this when she was complaining about a working professional. I asked her if she felt she would be treated with world class care from a doctor who used ChatGPT to pass biology and she sat with that question for a very long time. Her response,
High-stakes testing with proctored exams (F2F or online) is the only solution to this. I teach financial analysts, our concerns are the same. Students can’t do analysis without understanding the WHY of what they’re tasked with. They have to proof processes and prototype
Yo! I enjoyed thread, Professor Jain! But assuming students are just copy-pasting from ChatGPT overlooks how they might use it to learn—studies show AI can enhance, not replace, skills like debugging. I challenge you to a test in your class. With an example, like you stated,
this is a good take for exactly right now, but once context windows scale up to accommodate the largest of scopes, what’s next? eventually it will be the teachers who must adapt, not the students
My son is first year poli sci. Prof (as you say) provides scaffolding and project comprehension before the next piece is given. If turned in works hits 40% or higher in Ai detector, 10% penalty. If under 5% detection, 10% bonus. All essays over 20% of mark are in class open book
📚 3AM study vibes: Your coffee's cold, YouTube tutorials are running out, and that exam is in checks time oh no... Every student's late-night reality: 📝 "Just one more chapter" (x50) 🌙 Those weird study position experiments ☕️ The endless coffee runs Novis Series, gets your
Image
Why not lean into this and just teach kids how to code with AI. I want all our coders to embrace AI tooling in the real world ... Give them these skills on school.
Due to illness, I hosted online midterm exams. 60-80% of students in my five courses range from Brilliant to Genius. Stats students completed their 75-minute, 65-question exams with +95% accuracy in 20-30 min. A freshman in my 100-level course: 90 questions, 97% in 12 minutes.
GIF
Would it be possible that with AI a lot of people actually do not need to learn databases to unleash his/her productivity or creativity. So maybe those students actually don’t need to be in the class in the first place. The folks who are really interested in the inner working of
Give them a pop quiz on Inner & Outer Joins. Have them re-create an Index from scratch and track the performance enhancements and/or declines on the return. Make it partially Web-Cammed so they don't have the opportunity to use any of the AIs for a crutch.
My last 2 interns spent more time trying to get chatgpt and co-pilot to write things that they didn’t even complete their projects. Sometimes you gotta sift through legacy stuff and data specific issues that AI is not going to help much with
I’m currently taking a information technology class where one of the weekly assignments is to answer 3 questions based on the book, then ask your LLM of choice the same 3 questions and write out what your thoughts on each reply. Really interesting way to integrate new technology
When I took an advanced DB class in college, we had no quizzes, no tests, no midterms. Just one project for the semester. Simply … the professor provided a database, and a complex query. Our job was to rewrite that query in such a way that the SQL optimizer would DEoptimize
Ouroboros Crypsinous for Privacy This variant of Ouroboros is being developed to enable private transactions, a lesser-known aspect of Cardano’s privacy ambitions. Consensus and Staking Energy Efficiency Milestone Cardano’s PoS system consumes just 6 GWh annually (2021
GIF
Great thread, if you were serious about learning the fundamentals of coding self taught, what would you say is the most efficient route I.e. free online courses, websites, etc.?
i've thought about this as well, wonder if embracing llm's and pushing the complexity of the course and projects to the limits of these models is the way to go
Completely agree. Took a course at Harvard last year where chatGPT, Gemini, etc were not allowed. You were allowed to use DuckAI. It made a huge difference. Duck is a front end to ChatGPT but changed to not give you the answer. They call it a pedagogical AI. Forces you to do
I'm a chemical engineer and the same thing happened to ChemE skills when computers got powerful enough to solve the equations for us. We used to have to design a whole process by solving massive systems of linear equations. Nobody can do that any more, but we are all better
Databases is the class you want to pay attention in kids- I still use what I learned in that class every day and can’t say that about many others!
Was a C++ prof for years, and was an industry analyst when GenAI went nuts. It is absolutely going to change coding - for the better. Just like we use libraries, GenAI offers partial solutions. Edu has the challenge of incorporating it effectively. S0eeds dev though 😃
IT Teams hate being asked “Are you down?” during outages. Empower your campus with StatusCast - Trusted by top universities to keep students and staff informed and productive during outages. shorturl.at/EHJ02
Not sure what the issue is. But I hired quite a few Cal students since 2000 - not one was brilliant. And many didn't understand CS fundamentals such as networking, languages, databases. Their response to "What was your fav course" was..."Java". Perhaps it's the stupid
Academia is in for a reckoning that much is clear. Imo the business leaders hiring these students won’t care about the nuances of the thought processes. What AI is bad at now will be fixed in the next version.
Would you rather have an employee who is really good at doing arithmetic in their head, or 10 employees with calculators? For the vast majority of software jobs, AI will remove the requirement to think about code at all.
I’m not saying it’s a good thing over all, it’s not, but it’s a great thing for those kids who ARE doing the work (and are also AI savvy) and those who know the tech. Those are the people who will dominate.
Literally every single point you made will be entirely obsolete within 9 months as LLMs develop rapidly from the Nvidia GB200 rollout. You have no idea what you’re talking about nor how quickly everything you know and feel about the world will become obsolete.
How do you think code languages and databases will change with llms? Imho we’re just scratching the surface, as we’re using llms to code with languages built for humans. But will we need those programming languages at all when llms will be able to directly write machine
Kind of like knowing how to spell or write cursive is “outdated”, one may indeed be brilliant but if you can’t spell or hand write well or even effectively, you will be seen as a dullard by many. Also hard to gain trust and acknowledgement of accomplishment if one cannot