LinkedIn cofounder says students should expect tests to get harder to cheat on with ChatGPT — and to involve an AI examiner – He said oral exams would require students to develop greater knowledge rather than relying on AI.
https://www.businessinsider.com/linkedin-reid-hoffman-ai-education-college-exams-harder-cheat-2025-5
25 Comments
From the article
AI can make it easier to game traditional college assessments such as essays — so the way students are tested is likely to change, the LinkedIn cofounder [Reid Hoffman](https://www.businessinsider.com/reid-hoffman) says.
As a result, he says, students should expect college exams to become harder to fake their way through and to include an AI examiner.
“Wishing for the 1950s past is a bad mistake,” Hoffman said on a recent episode of his podcast “[Possible](https://www.youtube.com/watch?v=rl2IxQaxkGQ),” which he cohosts. “The fact that universities have not changed, and it’s like, ‘Well, but I already have my curriculum, and this is the way I’ve been teaching it for the last X decades,’ etc.”
Concerns regarding AI-driven [academic dishonesty](https://www.businessinsider.com/ai-cheating-colleges-plagiarism-chatgpt-professor-2024-9) have been on teachers’ minds since ChatGPT took off in late 2022. Plenty of students do use large language models as [homework help machines](https://www.businessinsider.com/ai-chatgpt-homework-cheating-machine-sam-altman-openai-2024-8), rather than slogging through the work themselves. Hoffman said the way students were [using AI to cut corners](https://www.businessinsider.com/how-college-students-use-ai-homework-study-2025-3) was circumventing the “whole point” of the educational system: learning.
Or hear me out, don’t make test online. Crazy concept i know. We had written comp sci tests in even back in 2019.
“…rather than relying on AI”
Given that AI often gives different answers to the same question, and its answers are wrong almost as often as it is right, anyone relying on AI is likely to be disappointed.
This all makes sense. When I was in high school you could write book reports almost solely on information from Wikipedia. Now they have the text checking software to combat that.
Any “cheat” you find, they’ll later try to stop it.
Why do business get to take the easy route with AI ? But not their consumers ?
oral exams is an ages old standard that is ai proof, degrees from any university that does not adopt that will turn into dust
Lmao, behold America’s CEOs.
Good Ole scantron and what we call paper and pencil should do the trick.
You do oral exams and all of a sudden people who suddenly cant pass will do amazing. All you have to do is make sure they have oral based instruction. Some people have reading disabilities, difficulties. Oral based learning and examination which can also involve hands on learning would be beneficial for some kids who are currently struggling
Except they won’t. The average teacher doesn’t have time to give an oral exam to 30 plus students per class. Specially when they probably have four classes a day or more
I think we have to fundamentally rethink the purpose of education. It feels that it’s present purpose is the separation of students into different ability levels rather than enabling students to optimise their contribution to society.
I think this is why so many fall through gaps or get ignored, if you don’t fit a very specific conformist concept of academic then you are passed over. However, who is to say that the skills future generations require will be the ability to optimise the use of AI.
We need to take a long hard look at the direction we are heading in.
The guy is afraid that AI is coming for him, too. No one is safe; even high-level executives are not safe. Look at Microsoft. Either way, the future will be bright. Not everyone will go on board, but a lot will either adapt or get left behind, and it looks like the LinkedIn co-founder will get left behind like others. Oh well, that is life and every era of invention.
Why would exam computers have online access or AI installed?
pretty sure paper and pencil exist. people who are online can schedule in person tests at testing centers.
if you’re not gonna validate your students then your degree and campus is worth nothing.
Ok. Then fund schools better so teachers can have fewer students and thus the time for all these suggestions.
There have been a number of articles coming out on students using Chat-GPT to skirt through college and the co-founder is obviously responding to that. Yet I’m not entirely convinced by his ideas on multiple fronts.
I agree that partially going back to oral exams could potentially adress the issue, but it’s not a panacea. For one, oral exams come with their own problems. They can be good for the testing of reproduction of certain concepts, definitions and theories but there simply isn’t too much room for theoretical depth due to the nature of a conversation and the time constraints. Especially when it comes to stimulating people’s own critical thinking capabilities and the creation of new knowledge, I can’t see it match essays and papers as a format. One might say you could try to cram an essay like question into an oral presentation, but if you divide the question into sub-questions with their own prompts, you can still off-load the critical thinking part to generative AI.
It’s also the reason I don’t buy his example of theses defenses. Yes, it puts you on the spot and leaves very little room for spontaneous bullshitting, but it’s still a defense of a written argument.
I also suspect there’s a problem with his example of using AI as a means to show what not do. In this case, it would have to be very clear that the bad examples were generated by AI, assuming he means bad examples of genuine prompts to AI, not just command AI to generate bad examples to emphasise the intrinsic value of doing work yourself. The problem is that the entire point of genAI like ChatGPT is that it is designed to be as convincingly human as possible. If it wasn’t, people would less likely to trust it and use it. But it’s very hard to definitively proof that an essay was written by AI. Teachers can usually tell intuitively, but intuition alone isn’t enough to make a case for plagiarism.
I also find the “AI-examiner” example frustrating for its lack of elaboration. “AI” is very susceptible to overinflation. What does he mean? An embodied examiner sounds like a fantasy and I also can’t imagine he means gen AI written paper feedback. For one, I’d be concerned with quality because hallucinations are inherent in how AI works and because feedback by statistical correlation of words isn’t the same as coceptual scrutiny (in philosophy papers for example). Second, Americans would be paying tens of thousands of dollars for this nonsense. You’d just have a conversation between an essay written by AI and feedback written by AI.
I agree with Zizek that this scenario is a potential blessing in disguise: “let the computers talk to each other and now we are free to learn” if it weren’t for the systematic constraints in which the education system operates. Universities are simply degree mills which hand out required stamps for the labor market. It’s not in the least surprising that students resort to gen AI. I’ve also seen lots of academics on Twitter suspect there is more going than just students using it. Likely, university governors and administratiors have been bought in too by educational consultants and start-ups marketing their newest bling with the idea that it will make their degree mill operations more efficient.
This has been the case since the late 1900s.
Same problem as the number 2 pencil on a scantron era.
Whether its AI and an essay or ABCD multiple choice, the quality of tests has gone down and the ability of students to wiggle through a test without knowing the material exists.
Which is why the gold standard of a test is going to be a combination of “write out your thoughts” and “sit down and talk to me. Explain this to me face to face”
But of course that higher standard of testing requires a lot more time and effort for teachers/professors, and there often just isn’t enough work hours to handle it
How about the AI being used to weed out millions of excellent candidates before they ever get past stage 0?
He’s projecting his fears on other people. Look buddy you’re going to get replaced to. So tighten up your bootstraps and get on with the program
Make school impossible to graduate. Then if you do graduate you can be employed to use AI.
socratic testing was always superior, funny how things evolve i mean devolve
My first degree had zero take home exams. We did mid-terms at mid-term inside an exam room at the university. We did finals at end of semester inside an exam room at the university. We were allowed a calculator from the approved list. The exam was a sheet of paper, the professor included a few constants or equations if they thought they didn’t need to be memorised. Exams were worth 90% of total. Take home papers were worth 10% total (because they knew we would chest a bit).
Is that just not possible anymore? Did the buildings all fall down or something??
If people would put in the same level of effort into actually learning the material as they do to try and cheat the test, they wouldn’t have to cheat the test.
Or we could reevaluate the education industry…these are the same people who said everyone needed to memorize equations because you wouldn’t always have a calculator.
AI is going to help kids at home so much. I can’t tell you how many times I had homework with no idea what to do and parents were no help. At least if you are getting answers from AI you’re at least (hopefully) reading it once instead of banging your head against the wall
If you can cheat on a test using chatGPT then it is something that doesn’t need to be tested.