Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Writing toy software is a joy

blog.jsbarretto.com

Post reply on HN
Print view View original

171–180 of 318 posts

Re: Writing toy software is a joy

#171

Post by kevinsync »

joshdavham wrote: post #170
dakiol wrote: post #39

> Perhaps you’re a user of LLMs. I get it, they’re neat tools. They’re useful for certain kinds of learning. But I might suggest resisting the temptation to use them for projects like this. Knowledge is not supposed to be fed to you on a plate Am I the only one using LLMs as if they were a search engine? So before LLMs I was searching on Google things like "pros cons mysql mongodb". I would read the official document…

> Am I the only one using LLMs as if they were a search engine? Nope, you're not alone as I also do this. I'm also not using any AI IDE's (yet). It's funny, I actually recently failed a live technical interview where I was using my LLM of choice to answer Google-like queries instead of an IDE like cursor. The interviewer told me that he had actually never seen anyone use AI for coding like that before. Up to that poi…

Not at all, I've been doing this with ChatGPT and Claude for a long time. I only recently (last couple weeks) started playing around with Claude Code on command line (not in an IDE). I didn't like Cursor very much. YMMV

Re: Writing toy software is a joy

#172

Post by jay_kyburz »

amunozo wrote: post #48
dakiol wrote: post #39

> Perhaps you’re a user of LLMs. I get it, they’re neat tools. They’re useful for certain kinds of learning. But I might suggest resisting the temptation to use them for projects like this. Knowledge is not supposed to be fed to you on a plate Am I the only one using LLMs as if they were a search engine? So before LLMs I was searching on Google things like "pros cons mysql mongodb". I would read the official document…

I tell to myself to use them as "teachers", not "interns", i.e. ask them questions to guide my process or look for the sources of knowledge needed to understand or do something, instead of asking them to get things done (except tedious, simple tasks).

Make sure to spend some time asking them questions on topics you already know a lot about. (I like to ask the AI about a game I developed called Neptune's Pride.)

A year ago the AI would just make up some completely random stuff. The current crop do a very good job, but still not 100% correct.

They are wrong enough that I would be wary of using them to "teach" me topics I don't know about.

Re: Writing toy software is a joy

#173

Post by bdunks »

snackbroken wrote: post #132
dakiol wrote: post #39

> Perhaps you’re a user of LLMs. I get it, they’re neat tools. They’re useful for certain kinds of learning. But I might suggest resisting the temptation to use them for projects like this. Knowledge is not supposed to be fed to you on a plate Am I the only one using LLMs as if they were a search engine? So before LLMs I was searching on Google things like "pros cons mysql mongodb". I would read the official document…

The few times I've used LLMs as question answering engines for anything moderately technical, they've given subtly-but-in-important-ways incorrect information such that taking them at face value would've likely lost me hours or days of pursuing something unworkable, even when I ask for references. Whether or not the "references" actually contain the information I'm asking for or merely something tangentially related…

Absolutely. And I’m finding the same with “agent” coding tools. With the ever increasing hype around Cursor I tried to give it a go this week. The first 5 minutes were impressive, when I sent a small trial ballon for a simple change.

But when asking for a full feature, I lost a full day trying to get it to stop chasing its tail. I’m still in the “pro” free trial period so it was using a frontier model.

This was for a Phoenix / Elixir project; which I realize is not as robustly in the training data as other languages and frameworks, but it was supposedly consuming the documentation, other reference code I’d linked in, and I’d connected the Tidewave MCP.

Regardless, in the morning with fresh eyes and a fresh cup of coffee, I reverted all the cursor changes and implemented the code myself in a couple hours.

Re: Writing toy software is a joy

#174

Post by leeoniya »

hbn wrote: post #121
leeoniya wrote: post #30

Earlier quoted context omitted.

> "by the way if you have friends who are expert programmers, I don't recommend asking them for help." AI is not an expert programmer [today]. and it doesnt take an expert programmer to arrive at that conclusion.

Call it whatever you want, it helps me get unstuck when my dependencies aren't getting along, or when I need to figure out what's wrong with my usage of an API I'm not familiar with, or why my build won't startup based on the code and the vague error message, etc. I don't care if it's a human or a robot telling me, I just want to get unstuck and I've been getting unstuck more since the advent of LLMs than in the rest…

i am NOT saying they are not useful; they are. especially in domains you are unfamiliar with, or boilerplate, or when being used by an expert in the loop. but that is NOT the same thing as being an expert.

an expert can write 10 lines of code that will be hyper fast while an LLM will write 30 lines that are super slow. both "will work", but one will fall over under load and will be hard to maintain and the other will not.

Re: Writing toy software is a joy

#175

Post by Aeolun »

cyberpunk wrote: post #152
lillecarl wrote: post #111

Earlier quoted context omitted.

I do this. But the killer usecase for me is writing all boilerplate and implementing some half-working stuff keeps my attention on the issue which makes me able to complete more complex things. A recent example is when I implemented a (Kubernetes) CSI driver that makes /nix available in a container so you can run an empty image and skip a lot of infra to manage. I talked to it a bit and eventually it wrote a Nix deri…

i’ve stopped writing “real” code for the most part, i just bang out some pseudo code like: read all files in directory ending in .tmpl render these as go templates if any with kind: deployment add annotation blah: bar publish to local kubeapi using sa account foo and tell it to translate it to x lang. so i control the logic, it handles the syntax. asking it to solve problems for you never seems to really work, but it…

It can solve problems, as long as they’re practical, or things done before.

Re: Writing toy software is a joy

#176

Post by Aeolun »

roughly wrote: post #115
zffr wrote: post #112

Earlier quoted context omitted.

Out of curiosity, how did you end up with 6 months off between jobs? I would love the opportunity to take ~6 months off, but I feel scared that if I just quit, it might take me a longer time to find a job again.

I'm single, rent controlled, and was working at a FAANG-ish company. The fear was real, for sure, but it honestly made a huge, huge difference for both my mental health and my employability afterwards. I also got to spend some time figuring out what I actually wanted to be doing, so I had a much clearer picture of where my skills and interests aligned after, which helped the search as well.

It probably helps when a FAANG style salary can pay for literal months of rent every single month.

I don’t want to be single again, but the absolute lack of concern over costs was nice.

Re: Writing toy software is a joy

#177

Post by Lerc »

I added up a score for myself from that list giving me one point for something that more-or-less matches an item from this list and a half point for something vaguely similar. I got 15.5

The compiler, chess engine, and text-editor have been targets for years but I haven't quite gotten around to doing them.

I like the idea of the bitwise challenge.

I had been thinking that another challenge that I'd like to see (and have a go at) might be a plotPixel Challenge.

Make a game that does all rendering with

    PlotPixel(x : int, y : int, color:??? )
    NextFrame(offScreen:boolean = true, clearNow:boolean =true, clearColor:??? =0)
Unsure on byte for 256 indexed color or 32 bit for setting 24 bit color from ARGB as r= rA+R, g=gA+G, b=b*A+B.

It would be easy enough to test if a game obeyed the rules because a full video capture of the game should be possible by intercepting just those functions.

Perhaps a (generous) source code size constraint to stop people going too crazy. (maybe 64/128/256k source-code, zipped). Not for code golfing as much as to avoid people trying to squeeze Crysis though a straw.

Re: Writing toy software is a joy

#178

Post by foretop_yardarm »

purplesyringa wrote: post #101

I'm surprised by the number of bad takes on LLMs in this thread. LLMs spoon-feed you with information about how things are implemented. You are not supposed to know how everything works when you start these projects. You're supposed to try your best, inevitably fail, then research the topic and understand where you went wrong, then adjust your approach. If you know how everything works and just follow the tutorial, y…

A lot of people say if you don’t use LLMs then you will fall behind. I’m starting to think that not using them will be a significant advantage in the long run.

Re: Writing toy software is a joy

#179

Post by Lerc »

alganet wrote: post #59
hn_throwaway_99 wrote: post #24

Earlier quoted context omitted.

I would just add that gen AI can actually be a fantastic aid when you're building some toy projects just for yourself. I am primarily a backend engineer, though I am capable on the front end, just not particularly fast. I am especially slow at the voodoo that is CSS. In the past, I would have been turned off from doing some end-to-end personal projects because I know I would have been bogged down just battling CSS fo…

A good reality check is: if a stranger asks you about a specific part of your toy project, would you be able to explain it? If you can't, it means there's something there you don't understand, and you lost an opportunity to learn. Of course, this is also true for reusing libraries and stuff. Within a job, what matters is the _team learning_, and it's roughly the same idea. At least one person in the team should be ab…

>A good reality check is: if a stranger asks you about a specific part of your toy project, would you be able to explain it?

I like that as a benchmark. I think it also works with LLMs too. I have had best results with AI generation of code when I pass that check. When using a LLM, the chatbot is the stranger.

In fact much of the frustration I have with using AIs seems to be for models tuned to the situation where the user does not have the ability to explain it specifically and it has to make broad assumptions. This may be the best result for a lowest common denominator, but it's frustrating when it provides a generic solution to a specific problem.

I can see specialised models diverging to focus on different levels of assumed knowledge and expectations. Perhaps to have modes of thinking, maybe as a more tone based MOE. Is anyone doing mixture of finetunes?

Re: Writing toy software is a joy

#180

Post by caseyohara »

whartung wrote: post #40
hn_throwaway_99 wrote: post #24

Earlier quoted context omitted.

I would just add that gen AI can actually be a fantastic aid when you're building some toy projects just for yourself. I am primarily a backend engineer, though I am capable on the front end, just not particularly fast. I am especially slow at the voodoo that is CSS. In the past, I would have been turned off from doing some end-to-end personal projects because I know I would have been bogged down just battling CSS fo…

It's a balance I find. On the one hand, the machines can spin up these bridges over unfamiliar code chasms. But on the other hand, you (at least I) have no investment in the code. No familiarity with it. No "skin in the game" when I can tell the AI to just redo it. One way around that is by using the "Chat bots as search engine", and then, rather than cut and pasting stuff into your project, type it in by hand. Even…

> rather than cut and pasting stuff into your project, type it in by hand

This type of mechanical copying is probably very important for learning and staying familiar with what it is doing.

Just last night I had ChatGPT help me with some algebra and geometry for a hobby project I'm working on. I'm rusty and got stumped so I asked the bot to unblock me. It helpfully walked through the entire simplification / substitution / solving process step-by-step, which I manually copied down using pencil and paper. Following along like that, mechanically, really helped it click for me. I plan to do more of that.

Post reply on HN