Hacker News Reader

Comment by CharlieDigital | Hacker News Reader

My 13 y/o was lamenting something the other day and it turns out that she had written a paper and pasted it into ChatGPT and asked it to grade the paper and it returned a "B+". She was using GPT to basically pre-check the quality of her paper before submitting it after having actually written it herself.

But it occurred to me that there are probably teachers out there who are taking the lazy route of using GPT to grade papers that students are submitting that have been written by GPT in the entirety.

There's some non-zero amount of GPT grading GPT in education.

View on HN

> it turns out that she had written a paper and pasted it into ChatGPT and asked it to grade the paper

that's one of the better uses of ChatGPT I've heard of

Sounds highly unreliable to me. I would be very surprised if ChatGPT was ever confrontational enough to give out a lower grade like D or F. It also wouldn't surprise me if the grade-letter it chooses tends to be off by one or more grade-letters just due to natural variance/subjectivity of grading and hallucinations.

GPT is blunt, critical, and highly opinionated if you tell it to be.

That's what I don't get about people on this forum. GENERATIONS of STEM professionals who have it engraved in their head that a computer will only do what you tell it to - and nothing more, and they still don't think to simply just tell GPT what lens they want it to view content in, or give it accurate and necessary directives. They think it'll handle that for them. Why?

This should be the group getting the best outputs of this technology, bar-none. I've never encountered a more pragmatic group of people, haha! I don't get where the disconnect is.

I agree. Sometimes I have some notes and thoughts I want to turn into a blog post, particularly pieces regarding philosophy or another such argumentation based field, and I'll ask an LLM to review and come up with counter examples. It does a very good job almost every time and really makes me think critically about the counter examples it gives me.

> I don't get where the disconnect is.

Some of us are not satisfied with 73% accuracy.

You're probably close to the same in everyday life - as are everyone around you, but I suppose that's valid.

> and they still don't think to simply just tell GPT what lens they want it to view content in, or give it accurate and necessary directives. They think it'll handle that for them. Why?

Projecting much? That doesn't describe me _at all_. I use ChatGPT and other LLM's frequently. I just legitimately don't think it would be good at the task of grading a paper with any sort of consistency. I don't even see why you would _need_ your paper graded for you to know it has flaws.

>"Projecting much? That doesn't describe me _at all_."

Didn't you just project, too? Lol. My comment is based on the writings I've seen here over the coming years about this technology.

That’s an improvement on what lazy teachers do now, grade the students current assignment based on their previous work or reputation or how much they like the student.

Tip for students: do well on your first assignment. The better teachers will give you the benefit of the doubt for mistakes on your future work. The bad teachers will just give you the same grade.

Having older siblings and friends with older siblings and resubmitting their essay is eye opening.

I use it this way for my own writing. I think this actually a good use. It allows for rapid feedback in away that doesn't require teacher time, and can help a writer grow. Of course LLM feedback should not be treated as gospel.

Props to your kid. Doing the work yourself and then asking for a critique (however flawed) is a way better approach than most kids at that age would take.

Did giving teachers guns and reducing their salaries not fix it?

Reply on news.ycombinator.com