Conversation

Notices

1. Embed this notice
   Christoph Becker (cbecker@hci.social)'s status on Wednesday, 21-Jun-2023 20:59:00 JST Christoph Becker

   Last week at one of the most progressive and thoughtful computing research communities I know and love, a peer-reviewed paper asked if #ChatGPT could "radically reimagine a new world." It illustrated something really important. 🧵 https://computingwithinlimits.org/2023/

   (This thread is not to dunk on the author or the paper. It is more the reverse: to illustrate how misleading metaphors and myths can pervade even the most progressive corners of computing and mislead us all.)

   • Embed this notice
     Christoph Becker (cbecker@hci.social)'s status on Wednesday, 21-Jun-2023 20:58:58 JST Christoph Becker

     in reply to

     Algorithms like #ChatGPT
     do not think,
     they do not dream,
     they do not lie,
     they don't pretend,
     they don't hallucinate, and
     they don't imagine.

     Because they are algorithms.

     Using these deeply human metaphors is a fundamental mistake that distorts what we can talk about when we talk about "#AI". Unfortunately it's all over the media and academic research too. https://doi.org/10.7551/mitpress/14668.003.0008

   • Embed this notice
     Christoph Becker (cbecker@hci.social)'s status on Wednesday, 21-Jun-2023 20:58:59 JST Christoph Becker

     in reply to

     Can #ChapGPT reimagine our world, radically or not?

     NO IT CAN'T.

     It's a #%$# language model #LLM. A stochastic parrot: https://doi.org/10.1145/3442188.3445922. It does not "imagine" anything. It generates plausible strings of words that are statistically likely to follow after the prompt, based on the training data. #AI

     pettter repeated this.