Conversation

Notices

1. Embed this notice
   Krista, Darth Moose Shark (grrrr_shark@supervolcano.angryshark.eu)'s status on Sunday, 29-Sep-2024 08:33:05 JST Krista, Darth Moose Shark

   @varx Yeah.

   The hard parts are understanding what the problem is and its constraints (often the hardest part), and figuring out how a solution within those constraints is structured, what it looks like, etc. Implementation can be difficult for complex solutions, sure, but its never about the actual coding. It's about figuring out all of the corners of the problem and solution. And by the time you have done that, asking an AI bot what it predicts such a solution would look like takes more work than translating it to code yourself.

   I have no issues with AI being used for language cleanup and summarisation if there is some human validation of the output - that is a reasonable task. My neurodiverse sometimes asks for summarisations of longer texts I send so she can understand the summary, and if she understands that, sends it to me and asks me if it is what I meant.

   None of it is worth the gigantic existential costs of LLM-generative AI that are happening now, but I can acknowledge this much at least 🤣

   • Embed this notice
     Krista, Darth Moose Shark (grrrr_shark@supervolcano.angryshark.eu)'s status on Sunday, 29-Sep-2024 08:33:02 JST Krista, Darth Moose Shark

     in reply to

     • Jeff Grigg

     @JeffGrigg @varx I think GenAI pollutes the already disastrously untethered information space and is driving us towards resource doom for no societal gain. But again, "X worked for my use case" is a user's truth that no one else is in a position to judge.

   • Embed this notice
     Krista, Darth Moose Shark (grrrr_shark@supervolcano.angryshark.eu)'s status on Sunday, 29-Sep-2024 08:33:03 JST Krista, Darth Moose Shark

     in reply to

     • Jeff Grigg

     @JeffGrigg @varx I am not stating the case that LLMs are good at anything. I am saying my daughter found it useful in a case where she could verify the output.

     Would I rely on it for anything? No.

     But if it is someone else's lived experience that it summarised what they were trying to say accurately and more clearly, who am I to argue with that?

     I used to work in computational linguistics research. Trust me when I say I am no fan of any of this crap.

     But I am not going to tell someone else they didn't have the experience they had.

     GreenSkyOverMe (Monika) repeated this.
   • Embed this notice
     Jeff Grigg (jeffgrigg@mastodon.social)'s status on Sunday, 29-Sep-2024 08:33:04 JST Jeff Grigg

     in reply to

     @grrrr_shark @varx

     Actually, insightful investigation has shown that LLMs are *no good at summarization*.

     They do not and cannot understand the *meaning* of the text. So they can't pick out the most important parts.

     All they can do is show you the most frequently repeated or similar parts. And often, that's the *least* important stuff:

     It often takes lots of evidence to reach a conclusion. And you have to eliminate alternative theories.

     Volume is NOT importance!!!