AI ‘Cheating’ Is More Bewildering Than Professors Imagined | Hacker News Reader

AI ‘Cheating’ Is More Bewildering Than Professors Imagined

theatlantic.com

26 pointsgHeadphone47 comments

The best way, so far, to prove that it's not written by ChatGPT is to show history using Google Docs. It's probably going to be the new "show your work" that math teachers went through during the rise of calculators.

AI should be easily able to generate that.

This is really smart. Although I imagine if it was very widely used, someone would come up with a way for AI to generate a fake history

It is easily circumvented by typing out the AI response rather than just copy and pasting

If you're writing something from scratch, though, you'd expect numerous revisions/deletions over time instead of just writing it from top to bottom in one shot. Rearranging sentences, changing phrasing, starting with the body of the work and coming back to write the intro later with lots of [insert thesis statement here] placeholders.

An AI could be trained to detect whether the typing follows human patterns or not. Same as how "I'm not a Robot" clicks work.

Time not spent teaching, researching, or running a university. Sigh.

Though it's probably the best way for now, I don't think it will last. Just like there's a market for 'mouse jigglers' to keep online status in the workplace... there will be automations for sale that will take GenAI output as input, and mimic all the natural mistakes and progress a human would make in writing a document.

Sure, but not everyone will use them. Or know how.

Which brings up the point of relying on false locks as security

Funny! I made a mouse jiggler (actual hardware mouse with switch to enable/disable) 10 years ago and there was no market! I even had a full unproductivity suite, like software to send slack messages at a later time and a chair with a motor that would randomly spin it around to make people think you just stepped out.

I need more info about the chair. Lmao.

All I can think of is this chair seemingly spinning after all humans are dead from the next pandemic. Ahaha

It’s just a regular office chair, you press a button then get up and walk away. It will then wait between 4-38 minutes before turning at least 100 degrees. If you press the button X times, it would shutdown after X hours. It really works (as in people tend to remember the last position of the chair) if you leave a sweater or jacket or something on the back.

That's utterly brilliant!

Just have the client spit out what ChatGPT does and overwrite it over time. I mean naturally you'd edit it too. Seems easy to hack that.

But would it seem worthwhile?

Great question

- Prewrite essay with ChatGPT

- Spend 15 minutes writing an essay structure in Google Docs (with the essay you already have, also maybe this step could be done for you by ChatGPT)

- Gradually rewrite the essay you've already got in front of you, which can look like a normal writing process

Congratulations, you have now not only got a legit looking edit history, but also you've rewritten what ChatGPT spat out, thwarting other potential methods of detection.

At this point, you might as well skip steps 1 and 2 and just write the damn essay.

if this process occurred from student' childhood, this is the only way to write some formal things.

In the future, we'll bake bread by scraping the toppings off frozen pizza.

Isn't this pretty much an argument for most GPTed writing?

Taking a GPTed piece of content, cutting out all of the useless stuff, proofreading and polishing it takes just as long as doing it by hand from the beginning?

This assumes a knowledge on the subject they're writing about.

Spoiler: plenty of students using this won't have a clue. That's why they're using it.

Easier still:

- Prewrite essay with ChatGPT.

- Ask ChatGPT to rewrite the essay in the voice of someone 5 years younger.

- Submit the juvenile copy as your rough draft.

You are playing GAN against 100.000 people and computers.

You never win a game of GAN

The next phase is for AI to grade all these essays regardless of how they are produced. So you'll end up with some version of that scene from Real Genius where it's just machines talking to machines.

One solution to this is very old school: in-person exams, hand-written, with no access to devices.

AI grading has been in use for years.

https://www.npr.org/2018/06/30/624373367/more-states-opting-...

I know; I've done research on it. But I don't know it's had the sort of widespread on-the-fly use in individual college courses in the same way as its application in large-scale standardized tests.

"...in-person exams, hand-written, with no access to devices..."

Maybe for math and subjects of that nature, but do you think your average college student could write out a 1000 word essay, or longer, without their hand falling off?

I have been typing for so long that it is physically difficult for me to do that while trying to think of what to write. Maybe I am too basic...well, probably...but I would assume the same for many other keyboard jockeys...

Just today I attended a 4th grade parents orientation at a Silicon Valley school where they noted they will be placing new emphasis on handwriting and cursive, specifically in order to build up legibility and hand stamina specifically for extended written exams.

The teachers are anticipating a return to hand written testing at high school and college levels in the AI-assisted present/future, so they are preparing students accordingly. (Then we visited the coding class and STEAM classrooms. It's not a Luddite reaction, more of a practical response to witnessed trends.)

The grade school pendulum has already started to swing from Chromebooks back to bluebooks.

This is good news...

These are easily defeated with small devices in the ear and shoe to guide the human. These are popular in Asia.

This whole thing is moot because precisely those subjects in which AI is most applicable are the ones whose graduates will most likely have their jobs taken away by AI anyway.

Like, at least let AI earn you the grades while you're in school, if it's going to put you out of a job once you're out, ya know?

Repeat after me: Education is not just about acquiring skillsets for jobs

- Signed every technical employee that didn't apply "Reproductive cycle of frog" at their Engineering desk

Yeah, it's also for drinking, finding a husband, teaching you about variable interest rates, putting you in the social circles of people above your class and/or potential employers, and drinking.

I would (and have done so) take a HS grad who taught themselves to code as a hobby over a fresh CS grad 8 days a week. If you want to be an academic and get into certain fields of research, then yeah college is for you, but any profession where education matters does their own testing a certification because they too know that undergrad is a joke.

And to get out ahead of it, I graduated with two degrees, a 3.9 GPA, and no debt so I'm someone who has every reason to gas up college as totally worth it because I benefited from the faux prestige.

> take a HS grad who taught themselves to code as a hobby over a fresh CS grad 8 days a week.

And he'll write you a function where the week does have 8 days!

In all seriousness, this is great if you can get a good hire at that level who is motivated and cheap enough employees that it's worth it to spend a lot of time coaching him.

But unless you're shipping low complexity CRUD apps or are in a very low cost location, it's generally much more cost-effective to hire from serious CS programs.

https://archive.ph/c1bG3

The AI-generated essay will be the new baseline. If you can’t argue beyond what an AI can or will generate, you haven’t broken through the NPC threshold.

If that means no more papers whose thesis statement is "the theme of Hamlet is revenge" I'm all for it.

Right. My high school rhetoric teacher used to label essays as “garbage” for not following structure or convention, often even for straying from expected narrative, while I was only interested in forming a unique or strong perspective. We may have both been wrong. I think the future of “essay” writing will be sort of like patents in that they will have to first identify a problem with a well-established viewpoint, then propose a solution, and fully describe its novelty by citing the prior works that form the boundary of the idea.

Wow, this is so different from my own experience (university professor here). I made using generative AI a requirement in my classes. It led to some inventive work, though most students are still figuring it out. I didn't make it central to the class, but will in the future. That being said, I don't assign an essay-type assignments but even if I did, I wouldn't use an AI checker like those mentioned in the article. At the start of the semester in January, I was shocked at how few students had actively used any AI tools (maybe 10%). At the end of the semester in May, I was shocked at how few of my colleagues had (25%).

What do you teach?

Grading is irrelevant to work performance, exactly like any other metric we try to measure when evaluating software developers' performance.

The only sane alternative is demonstrating skills through completing projects. Hopefully we move towards more than that

Is definitely frustrating as a software developer. The amount of time that I've spent preparing to be graded and then subsequently graded amounts to maybe a fifth of my waking life. Then a couple of years ago I knocked out some software projects and got a job purely based on them... What a waste of life.

If it's any consolation at any sufficiently large company you'll be graded again based on arbitrary metrics b/c your manager (and their managers) have no idea what your actual impact is.

The modern incarnation of school is, after all, supposed to make you into a good employee.

Had a manager tell me I needed to be more like another engineer who had more commits. I checked out his commits and it was full of repairing typos in READMEs, capitalizing SQL queries, removing comments from code.

I didn’t work for that manager much longer

I was hoping this would be another cautionary article on how ai likes to cheat rather than cheating using ai.

Reply on news.ycombinator.com