Conversation

Notices

1. Embed this notice
   ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 19:42:02 JST ✧✦Catherine✦✧

   every single AI-for-coding experiment i've conducted ended up the same way: it was clearly faster, even if only a little bit, to do the same thing without the LLM in the middle https://mastodon.social/@janl@narrativ.es/115014692424751266

   • Embed this notice
     ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 19:47:50 JST ✧✦Catherine✦✧

     in reply to

     • Wulfy—Speaker to the machines

     @n_dimension i have not conducted those experiments

   • Embed this notice
     Wulfy—Speaker to the machines (n_dimension@infosec.exchange)'s status on Tuesday, 12-Aug-2025 19:47:51 JST Wulfy—Speaker to the machines

     in reply to

     @whitequark

     Even in scenarios where you had no prior knowledge of language/frameworks?

   • Embed this notice
     ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 20:01:21 JST ✧✦Catherine✦✧

     in reply to

     • Wulfy—Speaker to the machines

     @n_dimension as someone who takes pride in excellent craftsmanship i consider this a waste of time

   • Embed this notice
     Wulfy—Speaker to the machines (n_dimension@infosec.exchange)'s status on Tuesday, 12-Aug-2025 20:01:22 JST Wulfy—Speaker to the machines

     in reply to

     @whitequark

     In my experience, the utility of AI increases with the expansion beyond your comfort boundaries...

     ... On the other hand there is a goldilocks zone there, because if you go too far out of your comfort zone... You will not be able to distinguish fact from fiction.

     Case in point:
     The attached image is from a 3D Demo I threw together in about 2-3 hours. It's in 3D python based framework called Visier. Spinny 3D shit hosted on the web. I know NOTHING about Python, Visier or hosting 3D frameworks on servers.
     Had I to do it the hard way, it would take me 3 weeks minimum, not 3 hours.
     I think THAT is true power of #vibecoding.

   • Embed this notice
     ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 21:13:27 JST ✧✦Catherine✦✧

     in reply to

     • Jan Lehnardt :couchdb:
     • yopp

     @alex @janl i was using auto-complete mode and the amount of subtle mistakes it inserted meant it would consistently cost me more in debug+review cycles than it would cost me to pay slightly more attention in the first place

   • Embed this notice
     yopp (alex@feed.yopp.me)'s status on Tuesday, 12-Aug-2025 21:13:28 JST yopp

     in reply to

     • Jan Lehnardt :couchdb:

     @whitequark @janl in auto-complete mode, no surpries, it is somewhat usable, especially when it’s a lot of boilerplate. But as “coding agent” it’s, uh, /suboptimal/

   • Embed this notice
     ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 21:17:16 JST ✧✦Catherine✦✧

     in reply to

     • Jan Lehnardt :couchdb:
     • yopp
     • David Chisnall (*Now with 50% more sarcasm!*)

     @alex @david_chisnall @janl if you can automate predictions for something to the point where an LLM can do it semi-reliably, in almost every case you could, and i will argue should, define an abstraction that does it deterministically

   • Embed this notice
     David Chisnall (*Now with 50% more sarcasm!*) (david_chisnall@infosec.exchange)'s status on Tuesday, 12-Aug-2025 21:17:18 JST David Chisnall (*Now with 50% more sarcasm!*)

     in reply to

     • Jan Lehnardt :couchdb:
     • yopp

     @alex @whitequark @janl

     I read that a lot, but to me 'a lot of boilerplate' is code smell. A tool that makes it easy to write a lot of boilerplate is a bad thing because it removes incentives to remove the boilerplate.

     If you need to duplicate a lot of code across different applications, you've introduced some fragility. It's hard to change the underlying APIs because everyone has copied and pasted the same thing (with or without an LLM). That's a problem for long-term evolution of a set of APIs. An LLM here just makes it easy to ship things that make everyone downstream accumulate technical debt faster.

   • Embed this notice
     yopp (alex@feed.yopp.me)'s status on Tuesday, 12-Aug-2025 21:17:18 JST yopp

     in reply to

     • Jan Lehnardt :couchdb:
     • David Chisnall (*Now with 50% more sarcasm!*)

     @david_chisnall @whitequark @janl depends on your definition of boilerplate I guess.

     Say I have event driven system and I want isolate each even type in its own “container” (class, module, whatever). Consequently there will be a lot of structural similarity between “containers”, because they will expose same interface.

     Some languages are verbose by design like HTML. If you are building a form you’ll have to repeat same things over and over again, and it’s… okay? Overall all UI/graphics code is very verbose because you have to setup a lot of things. And when you take component approach you’ll end up with case above.

     LLM autocomplete predictions are just more context aware than plain autocomplete (intellisense or whatever) and can save you some time on typing, because it can just spit out pre-filled method call, with all arguments filled with values from the current context.

     No wonder that thingie made to auto-complete based on a context does decent job as context-aware auto-complete!

   • Embed this notice
     ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 21:19:36 JST ✧✦Catherine✦✧

     in reply to

     • Jan Lehnardt :couchdb:
     • yopp
     • David Chisnall (*Now with 50% more sarcasm!*)

     @alex @david_chisnall @janl people have been writing abstractions over HTML for almost as long as HTML existed. you can go ahead and use .jsx/.tsx in almost any environment today; the abstractions have won

   • Embed this notice
     Wulfy—Speaker to the machines (n_dimension@infosec.exchange)'s status on Tuesday, 12-Aug-2025 21:19:36 JST Wulfy—Speaker to the machines

     in reply to

     • Ignas Kiela

     @ignaloidas @whitequark

     I am not 'scared'.
     I have been a developer of a sort of another for 35 years.

     I know exactly what I am capable of.
     It took me 6 hours to learn how to back up and restore a WP site.
     As you get older, tasks get harder.

     When I was 21, I could pull a 72 hour coding marathon fueled by 20 liters of Pepsi Max and a couple of short naps.

     Today, I am lucky to get a script running in a day.

     I actually do not like reading HTML tutorials, they are written very poorly, contain poor examples, and as anyone who used #StackOverflow will testify, half the 'solutions' are wrong.

     With AI, you just ask "Given these parameters, and these outcomes, how do I do blah?"

     Work harder, not smarter.
     Wait... the other one.

   • Embed this notice
     ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 21:19:36 JST ✧✦Catherine✦✧

     in reply to

     • Ignas Kiela
     • Wulfy—Speaker to the machines

     @n_dimension @ignaloidas people say this shit about getting older but i'm significantly more productive today than when i was in my early 20s. i aim for a bigger scope, i have a higher bar for quality, and i finish projects earlier anyway

   • Embed this notice
     Ignas Kiela (ignaloidas@not.acu.lt)'s status on Tuesday, 12-Aug-2025 21:19:38 JST Ignas Kiela

     in reply to

     • Wulfy—Speaker to the machines

     @n_dimension@infosec.exchange @whitequark@mastodon.social I think ypu underestimate yourself and you could have done it in 3 hours.
     
     What has recent research about AI usage show is that you're less scared to tackle new stuff if you have AI. You can do it if you're not scared.

   • Embed this notice
     ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 21:38:59 JST ✧✦Catherine✦✧

     in reply to

     • Jan Lehnardt :couchdb:
     • yopp
     • David Chisnall (*Now with 50% more sarcasm!*)

     @alex @david_chisnall @janl once more, the moment you are reaching for a snippet is an excellent moment to consider whether it could have been a function

   • Embed this notice
     yopp (alex@feed.yopp.me)'s status on Tuesday, 12-Aug-2025 21:39:01 JST yopp

     in reply to

     • Jan Lehnardt :couchdb:
     • David Chisnall (*Now with 50% more sarcasm!*)

     @whitequark @david_chisnall @janl and to make jsx/tsx components you have to setup said components and all the code around it is kinda boilerplate. Hardly any IDE have enough snippets to automate that.

     But I totally agree that with LLM you trade speed for attention, especially if it’s non statically typed code so your IDE can’t catch bullshit on the spot.

     I’m not defending LLMs: they are overhyped and they don’t live to their promise of the utility. For me it’s somewhat time saving in some scenarios, but overall it’s meh. I don’t believe that current architecture can achieve anything other than ruining the knowledge storages we had before

   • Embed this notice
     CaveDave (engravecavedave@mastodon.social)'s status on Tuesday, 12-Aug-2025 21:50:43 JST CaveDave

     in reply to

     @whitequark yeah, but have you considered the benefits of not having to think? /jk

   • Embed this notice
     ✧✦Catherine✦✧ (whitequark@mastodon.social)'s status on Tuesday, 12-Aug-2025 22:44:47 JST ✧✦Catherine✦✧

     in reply to

     • doragasu

     @doragasu the experiments i ran were all on code so mind-numbingly boring i was procrastinating doing it because it made me feel like my brain is fossilizing

     i.e. perfect LLM fodder

   • Embed this notice
     doragasu (doragasu@mastodon.sdf.org)'s status on Tuesday, 12-Aug-2025 22:44:48 JST doragasu

     in reply to

     @whitequark Not defending AI, I think it should be burned in flames along with the companies training them, but I think you are one of the most competent hackers out there. If an AI would match your level, it would definitely be something.