lobsters score: 1 · tags: plt, vibecoding
I believe that PLs are for people. As long some people write code for other people with intent and care, then PL design will matter for the original author, for the reviewer, for future maintainers, for hackers and for users (on account of enjoying a program whose behavior was grasped by a person).
For a PL researcher, I imagine the thought of my systems no longer being exclusively exclusively or even primarily used by people quite disheartening.
But the more research we put into empowering people to use PLs with ease (less cognitive load, friendlier tooling, simple semantics, etc) the more attractive and exciting it will be to write code by hand. I am really excited to learn Zig for example because I see many refreshing ideas in it coming from Rust. Of course it will never reach the same level of effortlessness as using LLMs, but it is something that many people care still care about.
I also think that it’s interesting to think about whether it’s possible to have a clean separation between human authored code and LLM-generated code inside a code base, whether we can have language features that help with restricting the effects of running slop (e.g. limited capabilities), etc.
It seems that recent online discourse is increasingly critical of LLMs and already thinking of hopeful ways forward. We’re moving away from the phase of astonishment and panic. That’s quite pleasant to see. Although maybe I shouldn’t have been so online in the first place :P
score 1 · canonical
It seems that recent online discourse is increasingly critical of LLMs and already thinking of hopeful ways forward.
Where do you see this? My experience is a lot more ambivalent.
- HN is like the anti-Lobste.rs. The majority is still as elated about LLMs as before, with a few critical voices as a vocal minority.
- Lobste.rs was critical from the get go, with a couple exceptions. This largely remained the same, with the absolute number of critical comments still dominating the discourse, though I’d argue I’ve seen more pro-AI comments here, than I did half a year ago.
- At work it’s only management that’s (giving the impression of being) excited, but that doesn’t mean the average programmer isn’t using LLMs, they’re just not loud about it. Admittedly, I don’t know how all departments operate, but the teams I do have vision on are absolutely working with AI without being forced to do so.
- Programmer friends are generally a lot more curious about LLMs than earlier, even those not super deeply into the topic and have previously discarded them as a toy are now asking those who are into them what harnesses they should use with what models. Most of them still think it’s “slop”, but are regardless interested in using it.
These all taken together doesn’t necessarily give me the impression that people are increasingly against the tech, just that the initial doom/hype wave is over and people are sobering up a little. My assumption is that we’re slowly establishing a baseline after the catastrophic initial shake-up of these tools being introduced into the field.
score 1 · canonical
though I’d argue I’ve seen more pro-AI comments here, than I did half a year ago
I’d argue the same handful of boosters comment on every genAI post possible and you’ll find they do the same thing on HN non-stop. It’s just manufacturing consent/borderline astroturfing, and it looks like it’s working on you.
score 1 · canonical
I can’t stay much about the noise around LLMs on Lobsters in particular. I’ve only joined recently after all.
I was speaking of my experience with the overall internet, e.g. on YouTube where “I’m done coding with AI” by Brett Codes hit 1M views in early Aug and was positively received in general.
Though I admit I was lurking HN more often in the past, and the general madness in the comments over there might’ve influenced my impression that things have gotten better since I opted to get my tech news from here instead.
score 1 · canonical
There is a weird disconnect between the first part of the post, which suggests that we should de-emphasize applications and can now focus on “the beautiful ideas, the insights, the beauty”, and then the second half, which presents the following three examples:
- Getting a machine to generate a possibly-correct (who knows?) declarative description of a programming language from an implementation of the language.
- Helping Amazon verify their proprietary software faster by adding LLMs in their verification tools.
- Asking a machine to produce a declarative specification of a programming language in a problem domain that we don’t understand, because we do not allow ourselves to spend the time to learn it (but can still serve as an expert to help the system authors assess the resulting spec).
I am sure that being able to work on all of these faster than before can be exciting and satisfying – this is the main point of the article. But I fail to understand how they relate to the proposal of worrying less about applications/adoption and focusing more on things that are beautiful and insightful.
(I also think that the post is tone-deaf and weird as a follow-up on the TYPES discussion, but oh well, this is a discussion to have with the author.)
score 5 · canonical
Well I suppose this is a manifesto for treating programming language research as a pure science.
I’m rather more interested in knowing what factors make programming languages readable. In this new era there’s already a lot of code that humans only read and don’t write. I strongly suspect that the ideal programming language for this era will be easy to read and make no concessions to being easy to write - the LLM can and should grind on checkers and making things consistent, but as humans we’ll be reading the code, commenting it, and very lightly editing it for the llm to follow our lead.
score 1 · canonical
If that’s the play, why not have everything written in Python and have the LLMs build a transpiler to Rust or something performant? That is what people laud the tools for, right?
score 1 · canonical
Or something like mojo, sure.
I’m not convinced that Python is the nirvana of readability, even if the syntax is fairly clean. It’s got plenty of spookiness, global state and the like, and very limited ways of declaring contracts and boundaries and delimiting pure and impure code.
My guess is that features that were tried and fell away (or never went mainstream) for “programming in the large” because they were annoying to write and redundant will make a comeback. I know I want more information about how code fits together without trying to work it out bottom up every time.
score 1 · canonical