Hacker News Reader

Comment by hansvm | Hacker News Reader

> have to manually run the code in your head

If you're doing a good job, you have to do that anyway, or at least have enough of a spidey sense for broken code to know when to investigate and add an extra test case.

Something like 30% of the time at $WORK, interviewers report the candidate as having solved the problem when a closer inspection reveals UB, memory corruption, and other bullshit. The test cases pass, and I think that's part of the problem. You can't tune out and avoid deeply understanding the submission.

View on HN

I'm fine with paper tests as a tradeoff, but in the real world where I am paid to program, I am a strong advocate of advanced type checkers/compilers and other static code analysis tools because they can keep all the code "in their head." They can keep far, far more code "in their head" than even the most gifted programming savants. Literally as I type out a mistake, the type checker alerts me to it.

So in some theoretical hiring example, if I have to choose between two candidates: one which is really good at keeping the code in the head and the other who is less good at that, but who is very skilled at using programming tools like advanced type systems or formal verification tools, then I'd prefer the latter.

Of course in practice, I would of course take a student that I knew had passed their paper test with their own knowledge over a student that likely ran their digital test through an LLM.

> If you're doing a good job

I think the problem is that the grader has to run your code in their head. That's a whole different problem.

I would imagine every professor (or assistant) who graded any programming exam I wrote was doing just that and further expect them to let some things slide as long as the general direction was correct.

They can OCR them and run them for real.

This would be HTR (Handwritten Text Recognition) not OCR and HTR is a lot harder than OCR especially for modern scripts (i.e. student's scrawls). And at every error the reviewer now needs to check if it was due to an error in the code or due to a bad HTR somewhere in the code.

I'd say compared to just typing the code on a disconnected computer: Not worth it.

And then you can grade them on important syntactic issues that really matter, like tabs -vs- spaces.

do they? OCR has gotten pretty good in 2026.

Good, the grader should hone that skill at pace then.

The student has to submit one exam.

The grader has to grade N exams.

Which is literally what graduate students get paid to do.

How do you think English, philosophy, or history tests get graded?

Do you think every institution of higher education has a ready supply of grad students to just throw at problems that require manpower?

Or were you just assuming your experiences (large university, I'm guessing) are universal?

Yes. That's exactly what college is. If you don't have the resources to effectively individually evaluate a student's output, you shouldn't hold a graded class.

Then let me spell it out for you, since you seem to be unaware:

There are many, many colleges that have no grad students. Period. They are undergraduate colleges.

Expecting any college or university to just be able to scale up the amount of manpower required to grade their classes/exams by a large multiple is absurdly unrealistic. I assure you that most existing colleges do have plenty of resources to effectively evaluate individual students' outputs, if they don't have to assume that anything submitted digitally is written by LLMs.

That new requirement is a massive change to how they operate, and it's incredibly rude and short-sighted to be dismissive and contemptuous of institutions that have trouble adjusting to it.

These colleges certainly have English, Philosophy, Art, and History classes, all of which require papers to be submitted and the exams consist of essay questions.

That Computer Science is somehow a privileged subject where the exam grading must be automated is elitism of the highest degree.

Reply on news.ycombinator.com