one of us! one of us!

Which specific idea? I can think of quite a few bits of things that react has that come from academia.

see ur/web, state monad, a long line of @shriram.bsky.social's work

syntactically or expressivity wise (amount of code needed to express an idea)

And again, as a testing expert you (and similarly other PL people with different verification expertise) might be well suited to do this work.

Regardless, I'm broadly of the opinion that researchers should do what they want and not be pulled by things that don't excite them.

I think the challenge is demonstrating to people that they even *should* use these systems (instead of taking the path of least resistance which is asking thr LLM to generate the tests for you)

- a majority of LLM / Agentic systems work is being done outside the community
- formal reasoning with AI people have been ahead of the curve so that's one place we've been doing well

- Bug data systems (spark, map-reduce were not PL conf papers)
- ML (TVM, PyTorch published at OSDI, ASPLOS)
- Cloud systems (hydro is at VLDB)
- High-performance design (HW design, image processing etc. came from outside the community)

This is an interesting thread because my perspective on your work is that it's precisely the kind of thing we need to make LLM-assisted code generation better *now*.

Some folks in formal methods already realize this but: when code is cheap to generate, verification becomes the primary problem

I would say the benefit of academia is the ability to take long-term bets instead of waiting for things to take their proper course.

PL historically has been late to the party on many different trends; it takes 2-3 years to really get a grasp on the ideas in a new area and contribute back

You can reasonably argue whether a class on C programming (or build systems, tooling, validating programs, etc.) would be useful or not but it is a materially different curricular goal from what a required, sophomore-level systems courses do (and I would argue, should do)

As @shriram.bsky.social always points out, it's more important to look at what the goals of the curriculum is instead of the specific language it uses.

The goals of most these classes is to teach students systems abstractions (which might include C but touch ISA, circuits, OS abstractions) ... 1/2

I don't think of "manufacturer prowess" (compilation is not novel-y hard, it is a lot of engineering).

Mobile definitely laid the groundwork

I don't have strong evidence to think so. For example, RISC-V has been in development for quite some time but compiler support is not super strong.

Said differently: this remains a function of consumer demand and compiler engineering remains slow