...because reading and writing well-written prose is meaningful and enjoyable?
It feels like half the people here do not read or write in their free time, which would be understandable if this were not primarily a site for software engineers who write (sorta) as a job
It is funny how that's basically one of the core points the article makes -- and in fact the article paints Hacker News commentors specifically as people who don't see that kind of inherent value in craft and artistry -- but the AI-generated summaries those people are relying on have missed it completely.
I actually disagreed with that particular point made in the article, because I don't really see myself as somebody who sees value in craft and artistry, I just want effective code that works (which imho LLMs cannot create).
But after reading this comment section... I mean if enjoying well written prose counts as enjoying craft and artistry I guess I do then? Damn.
> because reading and writing well-written prose is meaningful and enjoyable?
This is not prose, it is exposition. It is perfectly valid to critique any expository essay, especially one of this length, for its density (or lack thereof) of substantive information.
Sometimes writing can both contain information and be beautiful? This article is charming and thoughtful. Its style may not be for everyone, but for me it really hit, I am thoroughly enjoying reading it. Its style gives me no problem calling it prose.
A person writing an essay on their own site doesn't need to have the information density of bus timetable.
I somewhat disagree that this is not prose? This didn't seem like a purely expository piece. Like if it were just a straightforward technical piece than yeah its way to long, it could have been a few sentences.
But this seemed like it bridges the gap between prose and an expository essay -it was doing both.
> prose and an expository essay -it was doing both.
Putting prose in an essay means there are more valid criticisms of a piece of writing, not fewer. If somebody is breakdancing and reciting the periodic table at the same time it’s ok if somebody notices if they skipped the lanthanides and actinides.
I’m a fan of blending the two! It’s just really really hard to do both well at the same time. My most recent example is Malcolm Harris’ history of Palo Alto, it is incredibly well-done.
That’s kind of the point that I was making. When you mash the two together, both lenses are valid critiques.
It’s an exponentially more difficult way to accomplish either goal because one reader will see it and think “this is a sixteen thousand word essay that says very little” and another will see it and think “what a wonderful story” and there’s nobody to adjudicate who is correct.
Like I posted “this is sixteen thousand words about how the author doesn’t really use language models but might one day” and some folks’ rebuttal is that they enjoyed reading it. Those are two completely unrelated things! It’s like if folks saw the cover of The Hobbit and thought “Hell yeah!” and then when they read “there and back again” thought “whoever wrote that was being unnecessarily reductive”
I am a Junior dev (graduated in 2022) and I am gainfully employed at the McGowan institute, earning an okay salary, with colleagues who actively use LLMs, but there is zero talk of firing me or laying me off due of LLMs. I personally avoid LLMs for most things other than:
1) Google search which actually works
2) translating MATLAB, which I have never learned (and probably won't ever)
There is a whole team of Junior devs, and actually on Friday I got an email asking if I could refer another junior/entry level. Granted it was for a 1 year contract, but still.
These posts make me feel like I’m the worst llm prompter in existence.
I’m using a mix of Gemini, grok, and gpt to translate some matlab into c++. It is kinda okay at its job but not great? I am rapidly reading Accelerated C++ to get to the point where I can throw the llm out the window. If it was python or Julia I wouldn’t be using an LLM at all bc I know those languages. AI is barely better than me at C++ because I’m halfway through my first ever book on it. What LLMs are these people using?
The code I’m translating isn’t even that complex - it runs analysis on ecg/ppg data to implement this one dude’s new diagnosis algorithm. The hard part was coming up with the algorithm, the code is simple. And the shit the LLM pours out works kinda okay but not really? I have to do hours of fix work on its output. I’m doing all the hard design work myself.
I fucking WISH I could only work on biotech and research and send the code to an LLM. But I can’t because they suck so I gotta learn how computer memory works so my C++ doesn’t eat up all my pc’s memory. What magical LLMs are yall using??? Please send them my way! I want a free llm therapist and a programmer! What world do you live in?? Let me in!
A lot of people are using Claude Code which many consider to be a noticeably better for coding than the other models.
I think also they tend to be generating non-C++ code where there are more guardrails and less footguns for LLMs to run into. Eg they're generating Javascript or Python or Rust where type systems and garbage collection eliminates entire classes of mistakes that LLMs can run into. I know you said you don't use it for Python because you know the language but even experienced Python devs still see value in LLM-generating Python code.
That’s funny bc I linked my post to a server I’m on and I also was told to use an agent.
My worry about an agent is I’m trying to translate the math with full fidelity and an agent might take liberties with the math rather than full accuracy. I’m already having issues with 0 to 1 indexing screwing up some of the algorithm.
I'm firing you for being unable to adequately commune with the machine spirit.
(But for real, a good test suite seems like a great place to start before letting an LLM run wild... or alternatively just do what you're doing. We definitely respect textbook-readers more than prompters!)
Maybe I’m just a silly lab engineer who doesn’t know how big companies work (i did work at a startup for a time but that’s even more anti meeting than a lab) but I feel like maybe if you have so many back to back meetings that you need to plan around them you should have fewer meetings?
Also why is there a hard end time rather than a maximum time? What is even going on? How are you getting work done?
At large companies, there's just a lot of coordination needed. Which means a lot of meetings typically. Probably more than there is needed but one of the points of a large company is having a critical mass of people pulling in roughly the same direction(s).
>also why is there a hard end time rather than a maximum time?
Depends on company, culture, and individuals. If you get through the purpose of the meeting, the meeting organizer may well say "Let's give everyone 20 minutes (or whatever) back." I do think that it's a worthwhile goal to default to <1 hour meetings.
In any case, agendas and thinking about how the meeting time should actually be spent is worthwhile.
At least at my multinational company, we have teams iny department that are 12 hours ahead of us. That makes the window of overlap time we have small, so we often have blocks of meetings during that overlap time that compete for time slots for staff that make higher level decisions.
I really love Zed bc it’s fast and doesn’t try to run the code for you. I really dislike the “play” feature of IDEs. I learned Java in eclipse and for the linear time don’t know how to run it from the terminal. It has all the editing and highlighting features of a full ide but makes you run it from the terminal. And if you have Zellij or another muxer set up, it runs it in the Zed terminal automatically.
