Conversation

Notices

1. Embed this notice
   Aether ??? (aether@poa.st)'s status on Wednesday, 02-Sep-2026 03:47:25 JST Aether ???
   The safest job from AI may be writing.
   
   https://muratbuffalo.blogspot.com/2026/08/the-safest-j...
   
   Lol, not if Vox Day has his way
   • Embed this notice
     T man :sex: :puffgiga: :puffpowerroll: (theorytoe@ak.kyaruc.moe)'s status on Wednesday, 02-Sep-2026 03:47:23 JST T man :sex: :puffgiga: :puffpowerroll:

     in reply to

     • Wanderer über dem Nebelmeer
     @WandererUber @Aether
     I will say LLMs have gotten good at writing *passable* code
     they can get x out the door with a little effort, but they arent usually the best solution
   • Embed this notice
     Wanderer über dem Nebelmeer (wandereruber@poa.st)'s status on Wednesday, 02-Sep-2026 03:47:24 JST Wanderer über dem Nebelmeer

     in reply to
     @Aether you know, I am generally really pro AI, pro vibe-coding even. I like it, it's fun, it's useful.
     
     But I am getting really tired of this
     >LLMs have gotten very good at generating code
     They have? Show me. Show me the good code they produce. Show me the before. Explain the problem. Show me their solution.
     Because from where I'm standing, I'm seeing people produce 2 pages of text for a skill that explicitly explains to these "smarter than a PhD" systems that they should check if this functionality is in the std lib or an existing dependency, instead of reinventing the wheel.
     
     And whenever I peer under the hood in my projects, I have no fucking idea what any of the code is supposed to do. There's comments everywhere that tell me nothing. And it doesn't do it anyways.
   • Embed this notice
     T man :sex: :puffgiga: :puffpowerroll: (theorytoe@ak.kyaruc.moe)'s status on Wednesday, 02-Sep-2026 03:47:49 JST T man :sex: :puffgiga: :puffpowerroll:

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     • Wanderer über dem Nebelmeer
     @WandererUber @Aether
     if you wan good code you have to yell at the llm a bit and explicitly point out the issues
   • Embed this notice
     Wanderer über dem Nebelmeer (wandereruber@poa.st)'s status on Wednesday, 02-Sep-2026 03:51:05 JST Wanderer über dem Nebelmeer

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     @theorytoe @Aether Rember that coding challenge? Muse Spark failed. After multiple tries. With screenshots and debugging capabilities.
     Muse spark scores 62 on aa coding index. Opus 5 has 68.
     T man :sex: :puffgiga: :puffpowerroll: likes this.
   • Embed this notice
     Wanderer über dem Nebelmeer (wandereruber@poa.st)'s status on Wednesday, 02-Sep-2026 04:05:48 JST Wanderer über dem Nebelmeer

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     @theorytoe @Aether I did that. Keep coming back to needing my own harness and benchmark suite though, to get better at getting better code
     T man :sex: :puffgiga: :puffpowerroll: likes this.
   • Embed this notice
     T man :sex: :puffgiga: :puffpowerroll: (theorytoe@ak.kyaruc.moe)'s status on Wednesday, 02-Sep-2026 06:20:06 JST T man :sex: :puffgiga: :puffpowerroll:

     in reply to

     • Wanderer über dem Nebelmeer
     @WandererUber @Aether
     tbh hot take but I really dont think your harness is going to be the determining factor for weather you get good code or not
     
     if you have good supporting tools (like linters and whatnot) that can do code analysis, as well as documents that point out what you want to see and or review guidelines, thats going to be a lot more useful than a harness imho
   • Embed this notice
     bajax (bajax@baj.ax)'s status on Wednesday, 02-Sep-2026 06:51:01 JST bajax

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     • Wanderer über dem Nebelmeer
     @theorytoe @Aether @WandererUber they can write *great* code as long as you can specify exactly what you want and keep the scope of the tasks you give it small enough that the solution is obvious.
     T man :sex: :puffgiga: :puffpowerroll: likes this.