"To better understand how the kernel community views Rust,
we collect the posts about writing Rust drivers from lwn and
ycombinator until 2023/08/05, and use Chatgpt to analyze
them."

I guess that's research in the 2020s... https://www.usenix.org/conference/atc24/presentation/li-hongyu

@corbet Oh great ape lord Rhesus above... *sad screeches*.

Sad cc: @lkundrak

@ljs @corbet @lkundrak how do I brachiate away from this sad reality?

@corbet "Awarded best paper!"

@vbabka @corbet @ljs what the fuck
bro
1.) deform the rest of the body and rob it of nutrition so that brain can do thinking
2.) abuse the machines until they appear to be thinking

very smart humans!

not worthy of being called great apes

@corbet "Empirical" and "ChatGPT" do not belong in the same room

@corbet Ew

@corbet The Research Questions just shows that the whole thing isn't serious at all

RQ2: does RFL live up to the hype?

What does it mean to "live up to the hype" in an academic paper?

@corbet "empirical" "chatgpt". urh urh

@corbet@social.kernel.org That's... not at all what the paper says. It's a bit overly negative in places, but it's actually quite optimistic.

@corbet@social.kernel.org ~~To be extra clear, they don't even do what you quote at all (unless I have REALLY missed something). They give a pretty even-handed, clinical overview and mostly look at statistics regarding code safety/quality/dev time/build artifact data.

What makes you believe this is ChatGPT horror show nonsense? What is the purpose of your post?~~

EDIT: I found the quote in the appendix. Well-hidden! Ctrl-F wasn't finding it for some reason.

@corbet@social.kernel.org Ahhh, I found what you were talking about. Apologies; that was nestled away quite sneakily in the appendix!

Still, the paper is overall good. Sentiment analysis with ChatGPT is a little questionable, but the rest of the paper is a fairly even handed analysis.

@corbet@social.kernel.org Looking at this a little more, I suspect this was actually a request from a reviewer. The appendix isn't referenced in the paper (as far as I can see) and more just exists to give some evidence that RFL is "controversial". The main content of the paper stands on its own. I do wish that reviewers would stop saying "just use ChatGPT" though...

Again, apologies for the confusion. The paper was quite good as I was reading and USENIX is quite a well-respected institution in security, so I had a bit of a knee-jerk reaction.