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

271–280 of 318 posts

Re: Writing toy software is a joy

#271

Post by Gormo »

ngriffiths wrote: post #128
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 use it the same way. The feeling is that I'm back in ~2010 when Googling stuff felt like a superpower. I could find anything back then. Of course, it didn't last long, and trying to Google now is an exercise in pain and frustration. Lots of people have complained about the various things Google and marketers have done to get there, idk, I just don't like how it works now. Top LLMs feel amazingly good at rapidly sur…

If you do a web search and find a random blog post full of spelling errors and surrounded by ads, you're not going to trust that at the same level as a Stack Overflow post with a hundred upvotes, or an article with a long comment thread on HN.

But an LLM digests everything, and then spits out information with the same level of detail, same terminology, and same presentation regardless of where it came from. It strips away a lot of the contextual metadata we use to weigh credibility and trust.

Sure, you can follow references from an LLM, but at that point you're just using it a fuzzier form of web search.

Re: Writing toy software is a joy

#273

Post by Gormo »

Viliam1234 wrote: post #44
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…

> Now with LLMs, I simply prompt the same with a little bit more of context "pros and cons of using mysql vs mongodb when storing photos. Link references". In near future, companies will probably be able to pay lots of money to have their products come up better in the comparison. LLMs are smart enough to make the result seem "organic" -- all verifiable information will be true and supported by references, it will on…

This already happens unintentionally, e.g. Wikipedia loops, where bad info on Wikipedia gets repeated elsewhere, and then the Wikipedia article gets updated to cite that source.

When LLM-generated content is pervasive everywhere, and the training data for LLMs is coming from the prior output of LLMs, we're going to be in for some fun. Validation and curation of information are soon going to be more important than they've ever been.

But I don't think there'll be too much intentional manipulation of LLMs, given how decentralized LLMs already are. It's going to be difficult enough getting consistency with valid info -- manipulating the entire ecosystem with deliberately contrived info is going to be very challenging.

Re: Writing toy software is a joy

#274

Post by Gormo »

dgb23 wrote: post #262
patrickmay wrote: post #208

Earlier quoted context omitted.

I hear a lot of people talk about LLMs writing the "boilerplate" and wonder why they haven't abstracted that away in the first place. Maybe my brain has been permanently altered by hacking Lisp.

I hear you. But removing boilerplate via abstraction (Lisp) is very different from generating it on demand (LLMs). The former is obviously qualitatively better. But it requires up front design, implementation testing etc. The latter is qualitatively insufficient, but it gets you there with very little effort plus some manual fixes.

> The latter is qualitatively insufficient, but it gets you there with very little effort plus some manual fixes.

I remember years ago, when I worked at a large PC OEM, I had a conversation with one of our quality managers -- if an updated business process consumes half the resources, but fails twice as often, have you improved your efficiency, or just broken even?

"Qualitatively insufficient, but gets you there" sounds like a contradiction in terms, assuming "there" is a well-defined end state you're trying to achieve within equally well-defined control limits.

Re: Writing toy software is a joy

#275

Post by ngriffiths »

Gormo wrote: post #271
ngriffiths wrote: post #128

Earlier quoted context omitted.

I use it the same way. The feeling is that I'm back in ~2010 when Googling stuff felt like a superpower. I could find anything back then. Of course, it didn't last long, and trying to Google now is an exercise in pain and frustration. Lots of people have complained about the various things Google and marketers have done to get there, idk, I just don't like how it works now. Top LLMs feel amazingly good at rapidly sur…

If you do a web search and find a random blog post full of spelling errors and surrounded by ads, you're not going to trust that at the same level as a Stack Overflow post with a hundred upvotes, or an article with a long comment thread on HN. But an LLM digests everything, and then spits out information with the same level of detail, same terminology, and same presentation regardless of where it came from. It strips…

I agree and am perfectly happy using it as fuzzier web search, because it works really well for me.

Finding references is often my main goal, but other times I just want some quick facts, in which case I'll be as thorough with checking as I would when reading a random blog with spelling errors.

Re: Writing toy software is a joy

#276

Post by ngriffiths »

ngriffiths wrote: post #275
Gormo wrote: post #271

Earlier quoted context omitted.

If you do a web search and find a random blog post full of spelling errors and surrounded by ads, you're not going to trust that at the same level as a Stack Overflow post with a hundred upvotes, or an article with a long comment thread on HN. But an LLM digests everything, and then spits out information with the same level of detail, same terminology, and same presentation regardless of where it came from. It strips…

I agree and am perfectly happy using it as fuzzier web search, because it works really well for me. Finding references is often my main goal, but other times I just want some quick facts, in which case I'll be as thorough with checking as I would when reading a random blog with spelling errors.

Actually, someone should design one that can pull in quotes. Like as a separate tool the LLM uses to quote that has guarantees, it's just a copy and can't be hallucinated. Then you could see the primary info when it's needed/asked for, similar to articles

Re: Writing toy software is a joy

#278

Post by alganet »

mrheosuper wrote: post #229
alganet wrote: post #59

Earlier quoted context omitted.

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…

unless you own/write the whole software stack(not the web stack), i doubt anyone can fully explain their project. For example, if in your project you have to write a file, and someone ask you how does the kernel cache file writting, that would not be an easy question.

You're missing the point and taking it to an extreme.

Let's say I write a toy project to log all sites I visit to a text file. The goal is to learn browser extensions, not learn kernel cache. There's a clear boundary, and anyone asking me about kernel cache would sound very weird. It's obviously a lot of layers of knowledge deep into the stuff I'm learning.

If my toy project is a beautiful web app, then CSS is _not_ several layers deep. It is in the front of it. It's a dependency I _need to learn_ in order to own my project.

If my toy project is a web app, and doesn't matter if it looks beautiful or not, then I want to spend my time away from CSS (either writing it by hand or interacting with an AI). I am fine with having it ugly.

Understanding these dependencies and layers of knowledge is part of the learning as well.

Again, this works for teams as well. There are certain things, at a certain level, that if you don't have someone in the team that knows it, leads to a precarious situation.

The fact that people don't recognize this, and mistake it for not made here syndrome and other nonsense, is actually scary. It means those recipes (don't reinvent the wheel, good is better than perfect, etc) became mantras that people repeat without actually understand why they exist, and those boundaries have become arbitrary and meaningless.

Re: Writing toy software is a joy

#279

Post by Gormo »

andhuman wrote: post #250
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…

In the past we used to copy code verbatim from magazines. You have to start somewhere right?

But your brain was the clipboard. That simple process of transcription was something that you couldn't avoid learning from even if you wanted to. You'd notice the connections between the commands you typed and things that happened when you ran the program even if you weren't trying to.

Things would start to click, and then you'd have those moments of curiosity about how the program might behave differently if you adjusted one particular line of code or changed a parameter, and you'd try it, which would usually provoke the next moment of curiosity.

This was how many of us learned how to write code in the first place. Pasting the output from an LLM into your source tree bypasses that process entirely -- it's not the same thing at all.

Re: Writing toy software is a joy

#280

Post by nocman »

Viliam1234 wrote: post #44
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…

> Now with LLMs, I simply prompt the same with a little bit more of context "pros and cons of using mysql vs mongodb when storing photos. Link references". In near future, companies will probably be able to pay lots of money to have their products come up better in the comparison. LLMs are smart enough to make the result seem "organic" -- all verifiable information will be true and supported by references, it will on…

> LLMs are smart enough to make the result seem "organic"

I would never describe the output I've seen from LLMs as "organic".

Post reply on HN