This interview https://youtu.be/oWOz2htozfI?si=qdQ0uZRoZOYeThOn
from 2 days ago with a top researcher from OpenAI directly addresses the bitter lesson argument and the importance of scaling for the history of their models.
> Some of the numbers that you saw from the number of students who receive failing grades were because we caught them (cheating) and prosecuted them and are sending their cases to the center for student conduct,” Garcia said. According to Garcia, nearly 30 students in CS 10 were caught cheating on take-home exams in spring 2026.
In my uni, rates of honor code violations in introductory CS classes were high even before AI. I was a section-leader for the CS106 series at Stanford, and the honor code violations were common. In 2015, ~20% of one intro class was suspected of an honor code violation [1]. Often, the CS department comprised the majority of honor code violations in a given quarter.
There are several reasons for this:
1. Cheating in CS is easier to detect. MOSS [2] (authored by CS professor Alex Aiken) is a very effective tool at detecting plagiarism in coding assignments. Personally I witnessed more honor-code violations in math problem sets, but there was no feasible way for professors to detect this.
2. Problems in programming assignments are (usually) very tangibly wrong. I can bullshit my way through an essay with shoddy research, I can hand-wave a proof that is definitely wrong but will probably garner at least some points. But when your program is crashing or not compiling, and the due date is approaching, it produces a very immediate and undeniable sense of failure and pressure to cheat. The thing is, many students would get a decent chunk of credit even for failing code, but this is not immediately obvious.
3. The ability to cheat is more available. Math problem sets tend to change quarter by quarter. It's basically impossible to cheat on a prose essay short of straight up paying someone to write it for you, or fabricating sources. But for CS classes, especially at prominent universities, there are plenty of solutions online. Much of it is people who aren't event at Stanford implementing the assignments for fun or self-learning, and sharing it with their peers. Which, to be clear, isn't unethical or bad - it's the responsibility of Stanford students to refrain from looking at those solutions. But nonetheless, it's a contributing factor.
> MOSS [2] (authored by CS professor Alex Aiken) is a very effective tool at detecting plagiarism
He apparently also makes (I would assume a satisfying amount of) money selling the same technology to law firms for copyright/patent analysis: https://www.similix.com
(I love these ultra minimal HTML sites, ex. https://www.hwaci.com (SQLite commercial licensing) for another example. It just has this subtle smugness, like you either don't need any new clients or virtually all of the market is your client.)
Is flunking kids the right reaction to catching them cheating? If it was before LLMs, is it still? I would love to be able to hold the line and throw the book at anyone who cheats, but after the dam has burst does it still help to try to hold the water back?
The whole situation sucks for both students and teachers. Teachers know that the knowledge they're going to great effort to convey isn't going anywhere. Or at least, it's landing in far fewer fertile brains than it used to. Students are squeezed because part of the university experience is being forced to adapt to an academic load, and as a result change yourself in ways that benefit you (or at least produce learning!) There have always been relief valves -- not just forms of cheating, but blowing off a study session by using game theory on your grade or going to a tutor or taking easier classes or extending your stay at the school. But now there's this huge giant relief valve in the form of a shiny LLM that is always available, particular at 3:45am when your project -- the one you've steadfastly refused to use AI on thus far -- is due the next day. The schools have tuned the pressure for the old set of options, and it's not clear that there's a new tuning that maintains anywhere near the old level of learning.
I guess my question is: of those students who were flunked for cheating, how many of them were learning despite their cheating? (And how about the students who were cheating but not caught?) Also, what levers are there to move more students towards learning even with the chatbots present?
I'm sure these questions are being debated. I know Garcia personally, and he is very invested in his students learning. The title of his Joy course is legit. So I'm sure the profs have ideas around this, though clearly not happy ones. Perhaps I'll ask him.
I believe it’s still a single section, so probably around 250 (at least that’s about what it was when I was there a long time ago). Compared to the 1000+ who take 61A.
And if cheating was triggered using AI detectors, was it real?
AI detectors are pretty mid in practice - they tend to have a lot of false positives for "B" students who are okay, but can still be struggled to be more coherent than AIs are. There are some specific triggers that AIs are way more likely to do than students, but a lot of AI detectors will trigger on this "almost there, but you're still struggling" level of essay writing that might get a B, B-.
I could expect the same might be true for CS students even though I haven't seen how AI detectors work for CS/math homework.
You'd be amazed at how many students we know are obviously cheating because the logs reveal that they copy pasted a long, complete answer within seconds of opening a problem for the first time, full of sophisticated code constructs that we didn't teach them, and lot's of nicely formatted comments. Sometimes they even copy/paste the entire GPT output and then format it down.
This has been my wife’s experience as a college math professor. Instead of code it’s extremely formal problems with way more steps than the student normally performs using notation never taught in class.
It’s not that students didn’t cheat before, LLMs have just lowered the bar so far many can’t complete a live test in a class that requires effort.
Public schools around me have done more than just turn a blind-eye to cheating. During COVID, one of our kids cheated in every subject, we showed the proof to the school and only one of the teachers had any consequence at all, and that was to give her a failing grade for the test we could prove she cheated on (quizzes and HW were ignored).
It was trivial to show she had plagiarized every single essay she wrote (just put any unusual sounding sentence in google with double quotes around it), and her English teacher said roughly: "Yeah, lots of kids are doing that these days, nothing we can do about it"
Right, thank you. Personally I think reading all the code that the AI produces is impossible and kind of defeats the purpose of using it. The key is to devise a structured way to interact with it (skills and similar) and use extensive testing along the way to verify the work at all steps.
SEA with node.js "works" for nearly arbitrarily general node code -- pretty much anything you can run with node. However you may have to put in substantial extra effort, e.g., using [1], and possibly more work (e.g., copying assets out or using a virtual file system).
reply