> A common question is: “how much are students using AI to cheat?” That’s hard to answer, especially as we don’t know the specific educational context where each of Claude’s responses is being used.

I built a popular product that helps teachers with this problem.

Yes, it's "hard to answer", but let's be honest... it's a very very widespread problem. I've talked to hundreds of teachers about this and it's a ubiquitous issue. For many students, it's literally "let me paste the assignment into ChatGPT and see what it spits out, change a few words and submit that".

I think the issue is that it's so tempting to lean on AI. I remember long nights struggling to implement complex data structures in CS classes. I'd work on something for an hour before I'd have an epiphany and figure out what was wrong. But that struggling was ultimately necessary to really learn the concepts. With AI, I can simply copy/paste my code and say "hey, what's wrong with this code?" and it'll often spot it (nevermind the fact that I can just ask ChatGPT "create a b-tree in C" and it'll do it). That's amazing in a sense, but also hurts the learning process.

After reading the whole article I still came away with the suspicion that this is a PR piece that is designed to head-off strict controls on LLM usage in education. There is a fundamental problem here beyond cheating (which is mentioned, to their credit, albeit little discussed). Some academic topics are only learned through sustained, even painful, sessions where attention has to be fully devoted, where the feeling of being "stuck" has to be endured, and where the brain is given space and time to do the real work of synthesizing, abstracting, and learning, or, in short, thinking. The prompt-chains where students are asking "show your work" and "explain" can be interpreted as the kind of back-and-forth that you'd hear between a student and a teacher, but they could also just be evidence of higher forms of "cheating". If students are not really working through the exercises at the end of each chapter, but instead offloading the task to an LLM, then we're going to have a serious competency issue. Nobody ever actually learns anything.

Even in self-study, where the solutions are at the back of the text, we've probably all had the temptation to give up and just flip to the answer. Anthropic would be more responsible to admit that the solution manual to every text ever made is now instantly and freely available. This has to fundamentally change pedagogy. No discipline is safe, not even those like music where you might think the end performance is the main thing (imagine a promising, even great, performer who cheats themselves in the education process by offloading any difficult work in their music theory class to an AI, coming away learning essentially nothing).

P.S. There is also the issue of grading on a curve in the current "interim" period where this is all new. Assume a lazy professor, or one refusing to adopt any new kind of teaching/grading method: the "honest" students have no incentive to do it the hard way when half the class is going to cheat.

I feel like Anthropic has an incentive to minimize how much students use LLMs to write their papers for them.

In the article, I guess this would be buried in

> Students also frequently used Claude to provide technical explanations or solutions for academic assignments (33.5%)—working with AI to debug and fix errors in coding assignments, implement programming algorithms and data structures, and explain or solve mathematical problems.

"Write my essay" would be considered a "solution for academic assignment," but by only referring to it obliquely in that paragraph they don't really tell us the prevalence of it.

(I also wonder if students are smart, and may keep outright usage of LLMs to complete assignments on a separate, non-university account, not trusting that Anthropic will keep their conversations private from the university if asked.)

> Students primarily use AI systems for creating (using information to learn something new)

this is a smooth way to not say "cheat" in the first paragraph and to reframe creativity in a way that reflects positively on llm use. in fairness they then say

> This raises questions about ensuring students don’t offload critical cognitive tasks to AI systems.

and later they report

> nearly half (~47%) of student-AI conversations were Direct—that is, seeking answers or content with minimal engagement. Whereas many of these serve legitimate learning purposes (like asking conceptual questions or generating study guides), we did find concerning Direct conversation examples including: - Provide answers to machine learning multiple-choice questions - Provide direct answers to English language test questions - Rewrite marketing and business texts to avoid plagiarism detection

kudos for addressing this head on. the problem here, and the reason these are not likely to be democratizing but rather wedge technologies, is not that they make grading harder or violate principles of higher education but that they can disable people who might otherwise learn something

The writing is irrelevant. Who cares if students don't learn how to do it? Or if the magazines are all mostly generated a decade from now? All of that labor spent on writing wasn't really making economic sense.

The problem with that take is this: it was never about the act of writing. What we lose, if we cut humans out of the equation, is writing as a proxy for what actually matters, which is thinking.

You'll soon notice the downsides of not-thinking (at scale!) if you have a generation of students who weren't taught to exercise their thinking by writing.

I hope that more people come around to this way of seeing things. It seems like a problem that will be much easier to mitigate than to fix after the fact.

A little self-promo: I'm building a tool to help students and writers create proof that they have written something the good ol fashioned way. Check it out at https://itypedmypaper.com and let me know what you think!

How can I, as a student, avoid hindering my learning with language models?

I use Claude, a lot. I’ll upload the slides and ask questions. I’ve talked to Claude for hours trying to break down a problem. I think I’m learning more. But what I think might not be what’s happening.

In one of my machine learning classes, cheating is a huge issue. People are using LMs to answer multiple choice questions on quizzes that are on the computer. The professors somehow found out students would close their laptops without submitting, go out into the hallway, and use a LM on their phone to answer the questions. I’ve been doing worse in the class and chalked it up to it being grad level, but now I think it’s the cheating.

I would never do cheat like that, but when I’m stuck and use Claude for a hint on the HW am I loosing neurons? The other day I used Claude to check my work on a graded HW question (breaking down a binary packet) and it caught an error. I did it on my own before and developed some intuition but would I have learned more if I submitted that and felt the pain of losing points?

Interesting article, but I think it downplays the incidence of students using Claude as an alternative to building foundational skills. I could easily see conversations that they outline as "Collaborative" primarily being a user walking Claude through multi-part problems or asking it to produce justifications for answers that students add to assignments.

I think most people miss the bigger picture on the impact of AI on the learning process, especially in engineering disciplines.

Doing things that could be in principle automated by AI is still fundamentally valuable, because they bring two massive benefits:

- *Understanding what happens under the hood*: if you want to be an effective software engineer, you need to understand the whole stack. This is true of any engineering discipline really. Civil engineers take classes in fluid dynamics and material science classes although they will mostly apply pre-defined recipes on the job. You wouldn't be comfortable if the engineer who signed off on the blueprints of dam upstream of your house had no idea about the physics of concrete, hydrodynamic scour, etc.

- *Having fun*: there is nothing like the joy of discovering how things work, even though a perfectly fine abstraction that hides these details underneath. It is a huge part of the motivation for becoming an engineer. Even by assuming that Vibe Coding could develop into something that works, it would be a very tedious job.

When students use AI to do the hard work on their behalf, they miss out on those. We need to be extremely careful with this, as we might hurt a whole generation of students, both in terms of their performance and their love of technology.

I've used AI for one of the best studying experiences I've had in a long time:

1. Dump the whole textbook into Gemini, along with various syllabi/learning goals.

2. (Carefully) Prompt it to create Anki flashcards to meet each goal.

3. Use Anki (duh).

4. Dump the day's flashcards into a ChatGPT session, turn on voice mode, and ask it to quiz me.

Then I can go about my day answering questions. The best part is that if I don't understand something, or am having a hard time retaining some information, I can immediately ask it to explain - I can start a whole side tangent conversation deepening my understanding of the knowledge unit in the card, and then go right back to quizzing on the next card when I'm ready.

It feels like a learning superpower.

>Students also frequently used Claude to provide technical explanations or solutions for academic assignments (33.5%

The only thing I care about is the ratio between those two things and you decide to group them together in your report? Fuck that

My wife works at a European engineering university with students from all over the world and is often a thesis advisor for Masters students. She says that up until 2 years ago a lot of her time was spent on just proofreading and correcting the student's English. Now everybody writes 'perfect' English and all sound exactly the same in an obvious ChatGPT sort way. It is also obvious that they use AI when she asks them why they used a certain 'big' word or complicated sentence structure, and they just stare blankly and cannot answer.

To be clear the students almost certainly aren't using ChatGPT to write their thesis for them from scratch, but rather to edit and improve their bad first drafts.

My take: While AI tools can help with learning, the vast majority of students use it to avoid learning

No one seems to be talking about the fact that we need to change the definition of cheating.

People's careers are going to be filled with AI. College needs to prepare them for that reality, not to get jobs that are now extinct.

If they are never going to have to program without AI, what's the point in teaching them to do it? It's like expecting them to do arithmetic by hand. No one does.

For every class, teachers need to be asking themselves "is this class relevant" and "what are the learning goals in this class? Goals that they will still need, in a world with AI".

I'm looking forward to the next installment on this subject from Anthropic, namely "How University Teachers Use Claude".

How many teachers are offloading their teaching duties onto LLMs? Are they reading essays and annotating them by hand? If everything is submitted electronically, why not just dump 30 or 50 papers into a LLM queue for analysis, suggested comments for improvement, etc. while the instructor gets back to the research they care about? Is this 'cheating' too?

Then there's the use of LLMs to generate problem sets, test those problem sets for accuracy, come up with interesting essay questions and so on.

I think the only real solution will be to go back to in-person instruction with handwritten problem-solving and essay-writing in class with no electronic devices allowed. This is much more demanding of both the teachers and the students, but if the goal is quality educational programs, then that's what it will take.

I loved asking questions as a kid. To the point of annoying adults. I would have loved to sit and ask these AI questions about all kinds of interests when I was young.

It says STEM undergrad students are the primary beneficiaries of LLMs but Wolfram Alpha was already able to do the lion's share of most undergrad STEM homework 15 years ago.

This topic is also interesting to me because I have small children.

Currently, I view LLMs as huge enablers. They helped me create a side-project alongside my primary job, and they make development and almost anything related to knowledge work more interesting. I don't think they made me think less; rather, they made me think a lot more, work more, and absorb significantly more information. But I am a senior, motivated, curious, and skilled engineer with 15+ years of IT, Enterprise Networking, and Development experience.

There are a number of ways one can use this technology. You can use it as an enabler, or you can use it for cheating. The education system needs to adapt rapidly to address the challenges that are coming, which is often a significant issue (particularly in countries like Hungary). For example, consider an exam where you are allowed to use AI (similar to open-book exams), but the exam is designed in such a way that it is sufficiently difficult, so you can only solve it (even with AI assistance) if you possess deep and broad knowledge of the domain or topic. This is doable. Maybe the scoring system will be different, focusing not just on whether the solution works, but also on how elegant it is. Or, in the Creator domain, perhaps the focus will be on whether the output is sufficiently personal, stylish, or unique.

I tend to think current LLMs are more like tools and enablers. I believe that every area of the world will now experience a boom effect and accelerate exponentially.

When superintelligence arrives—and let's say it isn't sentient but just an expert system—humans will still need to chart the path forward and hopefully control it in such a way that it remains a tool, much like current LLMs.

So yes, education, broad knowledge, and experience are very important. We must teach our children to use this technology responsibly. Because of this acceleration, I don't think the age of AI will require less intelligent people. On the contrary, everything will likely become much more complex and abstract, because every knowledge worker (who wants to participate) will be empowered to do more, build more, and imagine more.

If I would start college today I would use all the models and chat assistants that are easily available. I would use Google and YouTube to learn concepts deeper. I would ask for subjects from previous years and talk with people from same and higher years.

When I was in college students were paying for homeworks solved by other students, teachers and so on.

In the article "Evaluating" is marked at 5.5% where creating is 39.8%. Students are still evaluating the answers.

My point is that just got easier to go in any direction. The distribution range is wider, is the mean changing?

I am currently in CS, Year 2. I'd argue that ~99% of all students use LLMs for cheating. The way I know this is that when our professor walked out during an exam, I looked around the room and saw everyone on ChatGPT. I have a feeling many of my peers don't really understand what LLMs are, beyond "question in, answer out".

While recognizing the material downsides of education in the time of AI, I envy serious students who now have access to these systems. As an engineering undergrad at a research-focused institution a couple decades ago, I had a few classes taught by professors who appeared entirely uninterested in whether their students were comprehending the material or not. I would have given a lot for the ability to ask a modern frontier LLM to explain a concept to me in a different way when the original breezed-through, "obvious" approach didn't connect with me.

I am surprised that business students are relatively low adopters: LLMs seem perfect for helping with presentations, etc, and business students are stereotypically practical-minded rather than motivated by love of the subject.

Perhaps Claude is disproportionately marketed to the STEM crowd, and the business students are doing the same stuff using ChatGPT.

They use an LLM to summarize the chats, which IMO makes the results as fundamentally unreliable as LLMs are. Maybe for an aggregate statistical analysis (for the purpose of...vibe-based product direction?) this is good enough, but if you were to use this to try to inform impactful policies, caveat emptor.

In my day, like (no exaggeration) 50 years ago, we were having the exact same conversation, but with pocket calculators playing the role of AI. Plus ca change...

I simply don't waste my time reading an AD as an article.

I take this as seriously as I would if McDonald's published articles about how much weight people lose eating at McDonald's.

As someone teaching at the university level, the goals of teaching are (in that order):

1. Get people interested in my topics and removing fears and/or preconceived notions about whether it is something for them or not

2. Teach students general principles and the ability to go deeper themselves when and if it is needed

3. Giving them the ability to apply the learned principles/material in situations they encounter

I think removing fear and sparking interest is a precondition for the other two. And if people are interested they want to understand it and then they use AI to answer questions they have instead of blindly letting it do the work.

And even before AI you would have students who thought they did themselves favours by going a learn-and-forget route or cheating. AI jusr makes it a little easier to do just that. But in any pressure situation, like a written assignment under supervision it will come to light anyways, whether someone knows their shit or not.

Now I have the luck that the topics I teach (electronics and media technology) are very applied anyways, so AI does not have a big impact as of now. Not being able to understand things isn't really an option when you have to use a mixing desk in a venue with a hundred people or when you have to set up a tripod without wrecking the 6000€ camera on top.

But I generally teach people who are in it for the interest and not for some prestige that comes with having a BA/MA. I can imagine this is quite different in other fields where people are in it for the money or the prestige.

I'd be very curious to know how these results would differ across other LLM providers and education levels.

My wife is a secondary school teacher (UK), teaching KS3, GCSE, and A level. She says that most of her students are using Snapchat LLM as their first port of call for stuff these days. Many of the students also talk about ChatGPT but she had never heard of Claude or Anthropic until I shared this article with her today.

My guess would be that usage is significantly higher across all subject, and that direct creation is also higher. I'd also assume that these habits will be carried with them into University over the coming years.

It would be great to see this as an annual piece, a bit like the StackOverflow survey. I can't imagine we'll ever see similar research being written up by companies like Snapchat but it would be fascinating to compare it.

I'm an undergrad at a T10 college. Walking through our library, I often notice about 30% of students have ChatGPT or Claude open on their screens.

In my circle, I can't name a single person who doesn't heavily use these tools for assignments.

What's fascinating, though, is that the most cracked CS students I know deliberately avoid using these tools for programming work. They understand the value in the struggle of solving technical problems themselves. Another interesting effect: many of these same students admit they now have more time for programming and learning they “care about” because they've automated their humanities, social sciences, and other major requirements using LLMs. They don't care enough about those non-major courses to worry about the learning they're sacrificing.

I use Claude as a Learning Assistant in my classes in Physics. I tell it the students are in an active learning environment and to respond to student questions by posing questions. I tell it to not give direct answers, but that it is okay to tell them when they are on the right track. I tell it that being socratic with questions that help focus the students on the fundamental questions is the best tack to take. It works reasonably well. I often use it in class to focus their thinking before they get together in groups to discuss problem solving strategies. In testing I have been unable to "jail break" Claude when I ask it to be a Learning Assistant, unlike ChatGPT which I was able to "jail break" and give students answers. A colleague said that what I am doing is like using AI to be an interactive way to get students to answer conceptual questions at the end of chapters, which they rarely do on their own. I have been happy using AI in this role.

If a student is passing your classes while using AI, I'm sorry your class is a joke.

Every class I sophomore on was open everything (except internet) and it still had a >50% failure rate.

I feel CS students, and to a lesser degree STEM in general, will always be more early adopters of advancements in computer technology.

They were the first to adopt digital wordprocessing, presentations, printing and now generative AI even though in essence all of these would have been disproportionately more hand in glove for the humanities on a purely functional level.

It's just a matter of comfortability with and interest in technology.

I’m about to graduate from a top business school with my MBA and it’s been wild seeing AI evolve over the last 2 years.

GPT3 was pretty ass - yet some students would look you dead in the eyes with that slop and claim it as their own. Fast forward to last year when I complimented a student on his writing and he had to stop me - “bro this is all just AI.”

I’ve used AI to help build out frameworks for essays and suggest possible topics and it’s been quite helpful. I prefer to do the writing myself because the AIs tend to take very bland positions. The AIs are also great at helping me flesh out my writing. I ask “does this make sense” and it tells me patiently where my writing falls off the wagon.

AI is a game changer in a big way. Total paradigm shift. It can now take you 90% of the way with 10% of the effort. Whether this is good or bad is beyond my pay grade. What I can say is that if you are not leveraging AI, you will fall behind those that are.

I'm curious why people think business is so underrepresented as a user group, especially since "analyzing" 30% of the Bloom Taxonomy results. My dual theories are:

- LLMs are good enough to zero or few-shot most business questions and assignments, so n.questions is low VS other tasks like writing a codebase.

- Form factor (biased here); maybe threads-only aren't best for business analysis?

So they can look very deeply into what their users do and have a lot of tooling to facilitate this.

They will likely sell some version of this "Clio" to managers, to make it easier for them to accept this very intimate insight into the businesses they manage.

I want to take an exception to the term cheat. Because it is only cheating the student in the end. I didn’t learn my times tables in elementary school. Sure, I can work out the answer to any multiplication problem, but that’s the point, I have to work it out. This slows me down compared to others who learned the patterns, where they can do the multiplication in their fast automatic cognitive system and possibly the downstream processing for what they need the multiplication for. I have to think through the problem. I only cheated myself.

With so much collaborative usage, I wonder how Claude group chats are not already a feature

an interesting area potentially missed (though acknowledged as out of scope) is how students might use LLMs for tasks related to early adulthood development. Successfully navigating post-secondary education involves more than academics; it requires developing crucial life skills like resilience, independence, social integration, and well-being management, all of which are foundational to academic persistence and success. Understanding if and how students leverage AI for these non-academic, developmental challenges could offer a more holistic picture of AI's role in student life and its indirect impact on their educational journey

Some students become better because of LLMs, some become worse.

It's like some people learn knowledge by TikTok, some just waste time on it.

I'm glad for AI, I was worried that future generation would overtake me, now I know they won't be able to learn anything

If you are doing remote learning and using AI to cheat your way through school you have obliterated any chance of fair competition. Cheaters can hide at home feeding homework and exams into AI, get a diploma that certifies all the cheating, then they go on to do the same at work where they feed work problems into an AI. Get paid to copy paste.

But I have a feeling that if it's that easy to cheat through life then its just as easy to eliminate that job being performed by a human and negate the need to worry about cheating. So I have a feeling it will work for only a very short amount of time.

Another feeling I have is mandatory in-person exams involving a locked down terminal presenting the user with a problem to solve. Might be a whole service industry waiting to be born - verify the human on the other end is real and competent. Of course, anything is corruptible. Weird future of rapidly diminishing trust.

What stops a student or anyone from creating a mashup of response and give back as something to teacher to check. Example feed output of Ollama to Chatgpt and that output to Google model and so on and then give final product to teacher for checking.

I don't think that can be caught.

professor here. i set up a website to host openwebui to use in my b-school courses (UG and grad). the only way i've found to get students to stop using it to cheat is to push them to use it until they learn for themselves that it doesn't answer everything correctly. this requires careful thoughtful assignment redesign. everytime i grade a submission with the hallmarks of ai-generation, i always find that it fails to cite content from the course and shows a lack of depth. so, i give them the grade they earn. so much hand wringing about using ai to cheat... just uphold the standards. if they are so low that ai can easily game them, that's on the instructor.

My BS detector went up to 11 as I was reading the article. Then I realized that "Education Report" was written by Anthropic itself. The article is a prime example of AI-washing.

> Students primarily use AI systems for creating...

> Direct conversations, where the user is looking to resolve their query as quickly as possible

Aka cheating.

This does not account for the ai usage used by students created by other companies such as openai etc...

AI bubble seems close to collapsing. God knows how many billions have been invested and we still don't have an actual use case for AI which is good for humanity.

I recently went back to school and got a look first hand how LLM's are used in classrooms.

1. During final exams, directly in front of professors: Check

2. During group projects, with potentially unaligned team-members: Check

3. By professors using "detection" selectively to target students based on prohibited groubds: Check

4. By professors for marking and feedback: Check.

And so, the problem is clearly the institutions. Because none of these are real problems unless you stopped giving a shit.

Good luck when you point out that your marked feedback is a hallucination and the professor targets you for pointing that out.

Simply match the student population one to one with AI and fit the curve as usual

> AI systems are no longer just specialized research tools: they’re everyday academic companions.

Oh, please, from the bottom of my heart as a teacher: go fuck yourselves.

I'm a professor at an R1 university teaching mostly graduate-level courses with substantive Python programming components.

On the one hand, I've caught some students red handed (ChatGPT generated their exact solution and they were utterly unable to explain the advanced Python that was in their solution) and had to award them 0s for assignments, which was heartbreaking. On the other, I was pleasantly surprised to find that most of my students are not using AI to generate wholesale their submissions for programming assignments--or at least, if they're doing so, they're putting in enough work to make it hard for me to tell, which is still something I'd count as work which gets them to think about code.

There is the more difficult matter, however, of using AI to work through small-scale problems, debug, or explain. On the view that it's kind of analogous to using StackOverflow, this semester I tried a generative AI policy where I give a high-level directive: you may use LLMs to debug or critique your code, but not to write new code. My motivation was that students are going to be using this tech anyway, so I might as well ask them to do it in a way that's as constructive for their learning process as possible. (And I explained exactly this motivation when introducing the policy, hoping that they would be invested enough in their own learning process to hear me.) While I still do end up getting code turned in that is "student-grade" enough that I'm fairly sure an LLM couldn't have generated it directly, I do wonder what the reality of how they really use these models is. And even if they followed the policy perfectly, it's unclear to me whether the learning experience was degraded by always having an easy and correct answer to any problem just a browser tab away.

Looking to the future, I admit I'm still a bit of an AI doomer when it comes to what it's going to do to the median person's cognitive faculties. The most able LLM users engage with them in a way that enhances rather than diminishes their unaided mind. But from what I've seen, the more average user tends to want to outsource thinking to the LLM in order to expend as little mental energy as possible. Will AI be so good in 10 years that most people won't need to really understand code with their unaided mind anymore? Maybe, I don't know. But in the short term I know it's very important, and I don't see how students can develop that skill if they're using LLMs as a constant crutch. I've often wondered if this is like what happened when writing was introduced, and capacity for memorization diminished as it became no longer necessary to memorize epic poetry and so on.

I typically have term projects as the centerpiece of the student's grade in my courses, but next year I think I'm going to start administering in-person midterms, as I fear that students might never internalize fundamentals otherwise.

I think there's ways for teachers to embrace AI in teaching.

Let AI generate a short novel. The student is tasked to read it and criticize what's wrong with it. This requires focus and advanced reading comprehension.

Show 4 AI-generated code solutions. Let the student explain which one is best and why.

Show 10 AI-generated images and let art students analyze flaws.

And so on.

See also https://www.spinellis.gr/blog/20250408/

"students must learn to avoid using unverified GenAI output. ... misuse of AI may also constitute academic fraud and violate their university’s code of conduct."

There's never mention of integrity or honor in these discussions. As if students are helpless against their own cheating. Cheating is shameful. Students should be ashamed to use AI to cheat. But nobody expects that from them for some reason.