Conversation

Notices

1. Embed this notice
   Eniko Fox (eniko@mastodon.gamedev.place)'s status on Wednesday, 21-May-2025 19:42:15 JST Eniko Fox

   AI is an eldritch horror. it is an incomprehensibly vast being but without physical form, not truly alive or aware of anything, though it consumes human knowledge and spits out a perversion of it. those who get too close to it are driven mad, and worship it as a deity that will lead them to salvation, even though it is not even aware of them and will lead them to destruction instead

   • Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Wednesday, 21-May-2025 21:16:44 JST Rich Felker

     in reply to

     • Phil

     @phil @eniko The only uses for is are where "producing information shaped drivel at scale" is a goal. That's mainly the domain of deceiving people/propaganda. It probably also "works" in domains like games where you need something that gives the appearance of something it's not, but folks rightly hate it there because it's cheap, regurgitive, and undermines actual creative work making much better things.

   • Embed this notice
     Phil (phil@fed.bajsicki.com)'s status on Wednesday, 21-May-2025 21:16:46 JST Phil

     in reply to

     @eniko@mastodon.gamedev.place I think that's a major exaggeration. There are valid uses for it; the hype isn't healthy, and training these things is destroying the planet for sure, but since we get free tools out of it, and inference costs less than a dollar a month (no, I don't use them all that much)... I don't think the problem is with users. It's corporations and regulators who should be held accountable for it.

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Wednesday, 21-May-2025 23:09:10 JST Rich Felker

     in reply to

     • Phil

     @phil @eniko It's well established both empirically and on sound theoretical basis that LLMs *cannot* summarize/review. But it's clear that you're already on the bandwagon and you're here to inject subtle (which becomes not-so-subtle as soon as you're challenged) pro-"AI" rhetoric into conversations, not to have your mind changed on something you're wrong about. 🙄

   • Embed this notice
     Phil (phil@fed.bajsicki.com)'s status on Wednesday, 21-May-2025 23:09:13 JST Phil

     in reply to

     • Rich Felker

     @dalias@hachyderm.io @eniko@mastodon.gamedev.place Ignorance is bliss.
     
     I've had great luck using LLMs (local) for reviewing my journals and notes. I even wrote a package for that: https://codeberg.org/bajsicki/gptel-org-tools
     
     I genuinely believe there's value in LLMs, provided that they're used in smart ways. They're dumb, bulky and stupid, but if you build massive walls and create a context in which the only possible option for them is to meet your requirements, it's quite possible.
     
     I shared a short snip of how I use this in the quoted thing, not sure if this works across fedi or not, so hard link also.
     
     https://fed.bajsicki.com/notes/a7uugdbodr
     
     My point here isn't that you're entirely wrong, because I do see that a lot of unscrupulous people do use them that way.
     
     My point is that for me, and many others, LLMs are a viable tool for self-reflection, and they are capable of speeding certain things up. E.g. I can have it review my sleep or exercise habits, or locate all the interactions I've had with people and remind me of the important bits.
     
     Ensuring the LLM works only on relevant data does make them more accurate. Sometimes they do miss things (like saying the novel I read is a manga), or get stuck in loops (especially around dates). This, for the most part, is easily fixed with a little bit of more specific instruction.
     
     RE: https://fed.bajsicki.com/notes/a7uugdbodr

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Wednesday, 21-May-2025 23:10:31 JST Rich Felker

     in reply to

     • Phil

     @phil @eniko "Spent far too long learning how to influence people... and then ditched that entirely.
     - Same for advertising. So many years wasted."

     Sounds like you haven't and you're still on that.

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Wednesday, 21-May-2025 23:15:09 JST Rich Felker

     in reply to

     • Phil

     @phil @eniko Yes. You got up in my mentions simping for "AI".

   • Embed this notice
     Phil (phil@fed.bajsicki.com)'s status on Wednesday, 21-May-2025 23:15:10 JST Phil

     in reply to

     • Rich Felker

     @dalias@hachyderm.io @eniko@mastodon.gamedev.place
     I'm not the one resorting to ad hominem, for the record.
     
     Have I done a single thing to anger or offend you?

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Wednesday, 21-May-2025 23:23:18 JST Rich Felker

     in reply to

     • Phil

     @phil @eniko I'm not offended by you running local models. What I object to is your insertion of gratuitous pro-"AI" narratives into public conversations about the topic with misleading claims about utility.

     As for the last sentence, projection much. 🤡

   • Embed this notice
     Phil (phil@fed.bajsicki.com)'s status on Wednesday, 21-May-2025 23:23:20 JST Phil

     in reply to

     • Rich Felker

     @dalias@hachyderm.io @eniko@mastodon.gamedev.place
     Not sure what simping is, and I don't believe I have ever referred to LLMs as 'AI'.
     
     It's a tool with a very, very narrow scope of applications. Just because people are abusing it, doesn't mean everyone is.
     
     I run local models, maybe for 5-10 minutes every few days. The idea that this offends you is genuinely funny to me.
     
     Every single thing I publish, I write myself (unless it's examples of the tool working, in which case it's obviously noted as such.)
     
     Your rage is best directed as regulators and corporations training these models - that's where all the power and environmental damage happens.
     
     For individual users, such as myself, the damage is roughly equivalent to leaving a lightbulb on for a few hours, or playing a video game for those same 5-10 minutes.
     
     In that perspective, your attitude is rather jarring, since you act as if I'm somehow responsible for what Google/ MS/ Elon/ OpenAI/ etc are doing.
     
     Almost like you're in a cult, really.

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Wednesday, 21-May-2025 23:27:20 JST Rich Felker

     in reply to

     • Phil

     @phil @eniko 🙄 🤡

   • Embed this notice
     Phil (phil@fed.bajsicki.com)'s status on Wednesday, 21-May-2025 23:27:21 JST Phil

     in reply to

     • Rich Felker

     @dalias@hachyderm.io @eniko@mastodon.gamedev.place
     I wouldn't agree it's misleading. I put in a bunch of time and effort to make this work for my specific use-case, and it does work for me.
     
     So, not misleading.
     
     But then my question is, if it's not a cult, why do you need it to be a lie?