Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.compilers > #3745 > unrolled thread

Why should students still study compilers?

Back to article view | Back to comp.compilers

Contents

  Why should students still study compilers? Michael Lehn <michael.lehn@uni-ulm.de> - 2026-08-25 08:46 +0200
    Re: Why should students still study compilers? bixbox <noreply@example.invalid> - 2026-08-27 07:48 -0400

#3745 — Why should students still study compilers?

As a mathematician, I can only comment on this as an outsider. But perhaps an
outside perspective has some value here.

Even though in mathematics (to my regret) it has become increasingly common to
motivate topics by explaining what they might be useful for later, most of
what we teach is about things that already exist. We teach theorems that have
already been proved. Exercises and exam questions have usually been asked
before in similar, and sometimes even identical, form.

Since the arrival of LLMs, there has been something close to panic among some
colleagues that students can now simply solve everything with ChatGPT. What
tends to be forgotten is that students could always copy solutions to homework
or take-home exams from other students without understanding them. They could
also memorize solutions to standard exam problems without understanding why
they work.

But there have also always been students who wanted to know _why_  something
works, either because they were curious from the beginning or because somebody
managed to make them curious. They wanted to understand how one might come up
with a proof rather than merely reproduce it. Those students gradually
developed their own ways of thinking. Later they were not only better at
dealing with known theorems and standard problems, but also at developing
creative ideas when confronted with problems they had never seen before.

Mathematics has done rather well with the idea that learning how existing
things work prepares you for challenges you do not yet know. That also means
not merely using tools — known theorems, for example — without having any idea
why they work.

Of course it is neither possible nor necessary to understand everything down
to the last detail. But I think it helps enormously to enjoy trying to
understand as much as possible, without constantly being distracted by the
question “what will this be useful for?” If you really understand something,
some useful consequence will often turn out to be a by-product — possibly one
that nobody could have predicted when you learned it.

I have no idea which problems LLMs (which I use myself) will be able to solve
better than I can in the future. But I suspect I will be able to use them much
better if I understand which problems I can delegate to them as part of
solving a larger problem that they cannot yet solve the way I can.

And coming back to compilers: I find it difficult to understand why one would
_not_  want to know how a compiler works. Quite apart from whether one will
ever write a compiler professionally, I think it is simply interesting to
understand how the thing that turns your program into something a machine can
execute actually works.


Michael Lehn

University of Ulm, Institute for Numerical Mathematics
Helmholtzstr. 20
D-89069 Ulm, Germany
Phone: (+49) 731 50-23534, Fax: (+49) 731 50-23548

[toc] | [next] | [standalone]

#3748

Michael Lehn <michael.lehn@uni-ulm.de> writes:

> But there have also always been students who wanted to know _why_  something
> works, either because they were curious from the beginning or because somebody
> managed to make them curious. They wanted to understand how one might come up
> with a proof rather than merely reproduce it. Those students gradually
> developed their own ways of thinking. Later they were not only better at
> dealing with known theorems and standard problems, but also at developing
> creative ideas when confronted with problems they had never seen before.

Which in my opinion is one of the advantage to use LLM. Having the
capability on your fingertips to dig deeper on the why, or even basic
more comprehensive explanation and the capability to expand my knowledge
very easily and the capability to compress the learning time.

> I have no idea which problems LLMs (which I use myself) will be able to solve
> better than I can in the future. But I suspect I will be able to use them much
> better if I understand which problems I can delegate to them as part of
> solving a larger problem that they cannot yet solve the way I can.

One amazing thing that I was able to do is to apply some formal
verification methodology to my day to day work because using LLM enable
me to explore and comprehend way more complex problem in a shorter time
and with less effort.

bix
[I suppose.  How do you know that the LLM got it right?  I'll say it did,
but that's not the same thing? -John]

[toc] | [prev] | [standalone]

Back to top | Article view | comp.compilers

csiph-web