Conversation

Notices

1. Embed this notice
   Collin Donnell (collin@ruby.social)'s status on Friday, 22-Sep-2023 09:50:40 JST Collin Donnell

   • Thomas 🔭🕹️

   @thomasfuchs comparing them to crypto seems nuts to me. LLMs are useful for real things by many thousands of people right now, even if they aren't useful for everything people want them to be. Crypto has existed for like 15 years and has never been shown to be useful for much of anything.

   • Embed this notice
     Stu (tehstu@hachyderm.io)'s status on Friday, 22-Sep-2023 12:46:42 JST Stu

     • Thomas 🔭🕹️

     @thomasfuchs every "they're good, actually" reply should include a brief explanation as to whether the poster accepts whatever utility they provide is likely stolen/scraped from someone else.

   • Embed this notice
     mypalmike (mypalmike@macaw.social)'s status on Friday, 22-Sep-2023 23:43:37 JST mypalmike

     in reply to

     • Thomas 🔭🕹️
     • Chris P. :trek_ds9_sisko:#1️⃣

     @b4ux1t3 @thomasfuchs And then sometimes it's well beyond that. It sometimes correctly predicts that you will add some new function/method and can implement it. Or I don't know a language well or an API. I just write a comment like "// call Mastodon server and get the vote data for the toot as JSON" and BAM, there's the code to do that. 5 seconds vs at least a few minutes for me. I suppose YMMV but I have found it a massive productivity boost and don't understand how others could possibly not.

   • Embed this notice
     mypalmike (mypalmike@macaw.social)'s status on Friday, 22-Sep-2023 23:43:38 JST mypalmike

     in reply to

     • Thomas 🔭🕹️
     • Chris P. :trek_ds9_sisko:#1️⃣

     @b4ux1t3 @thomasfuchs Referring to LLM for coding, I'm mostly just referring to Copilot, which doesn't involve asking a question, you just start writing code (or even just a comment about what the upcoming code is supposed to do) in VSCode and it very often generates a perfect line or block of code. And by perfect, I mean it's exactly what I would have typed, but instantaneous. Sure it misses sometimes, but it's obvious when it does.

   • Embed this notice
     Chris P. :trek_ds9_sisko:#1️⃣ (b4ux1t3@hachyderm.io)'s status on Friday, 22-Sep-2023 23:43:39 JST Chris P. :trek_ds9_sisko:#1️⃣

     in reply to

     • Thomas 🔭🕹️
     • mypalmike

     @mypalmike @thomasfuchs uh... Every project I've ever even considered applying them to?

     If I need to figure out some tricky bit of syntax, taking the time to lay out what I need in a way that a LLM can understand it makes me figure out what I need before even sending the question.

     That's called rubber ducking, and it isn't new.

     Unit tests? Source generation.

     Boilerplate? Source templates.

     The few times I've actually asked it meaningful questions, it confidently gave me incorrect syntax.

   • Embed this notice
     mypalmike (mypalmike@macaw.social)'s status on Friday, 22-Sep-2023 23:43:41 JST mypalmike

     • Thomas 🔭🕹️

     @thomasfuchs LLMs are amazing tools for a solo developer that are productivity multipliers. Not just code, which they are amazing at, but many other aspects of development.

     "Teach me about data management using Angular." Drill in on details in the answer.

     "Make a list of (real world data - e.g. uh... guitar manufacturers, headquarter cities, and founding dates) and format it as JSON." I'm gonna check the data of course.

     What projects have you worked on where LLMs failed to deliver?

   • Embed this notice
     SomeGadgetGuy (somegadgetguy@techhub.social)'s status on Saturday, 23-Sep-2023 01:04:59 JST SomeGadgetGuy

     • Thomas 🔭🕹️

     @thomasfuchs "I TOTALLY ALWAYS knew it was going to be a bust, and I definitely don't have a portfolio full of worthless crypto and NFTs either!"
     🙄

   • Embed this notice
     pax (pax@mastodon.com.pl)'s status on Saturday, 23-Sep-2023 01:16:56 JST pax

     • Thomas 🔭🕹️

     @thomasfuchs I hope not, as llm's are good for accessibility.

   • Embed this notice
     poetaster (poetaster@mastodon.gamedev.place)'s status on Sunday, 24-Sep-2023 06:36:06 JST poetaster

     • Thomas 🔭🕹️

     @thomasfuchs VisionEncoderDecoder models are cool, especially the ocr- free . But that's less the LLM stochastic parrot bit, more, well, classical.