Conversation

Notices

1. Embed this notice
   Anthony (abucci@buc.ci)'s status on Thursday, 30-Nov-2023 18:53:03 JST Anthony
   Add this to the list of real and present dangers in the current crop of #LLM / #AI models like #ChatGPT:
   Using this tactic, the researchers showed that there are large amounts of privately identifiable information (PII) in OpenAI’s large language models. They also showed that, on a public version of ChatGPT, the chatbot spit out large passages of text scraped verbatim from other places on the internet.
   From https://www.404media.co/google-researchers-attack-convinces-chatgpt-to-reveal-its-training-data/
   
   • pettter repeated this.
   • Embed this notice
     Anthony (abucci@buc.ci)'s status on Thursday, 30-Nov-2023 18:53:10 JST Anthony

     in reply to
     Given the scale of the training data sets used to train models like this, it is infeasible to ensure, in any reasonable sense, that the training data does not contain PII. Doing so would surely destroy any hopes of profitability the companies making these have. Thus, this is a new attack vector and a new externality that we're meant to simply accept collectively. Personally I don't recall being asked whether I'm OK with that...
     
     #LLM #AI #OpenAI #ChatGPT