Conversation

Notices

1. Embed this notice
   Carl T. Bergstrom (ct_bergstrom@fediscience.org)'s status on Thursday, 18-May-2023 03:35:44 JST Carl T. Bergstrom

   Low-hanging fruit for Bing Chat: right now, search queries yield retracted papers with no warning attached.

   I don't except LLMs to get this sort of thing right, especially if trained before the retraction, but hybrid systems like Bing Chat ought to.

   • Embed this notice
     DrYak (dryak@mstdn.science)'s status on Thursday, 18-May-2023 03:35:42 JST DrYak

     in reply to

     • Cory Doctorow

     @ct_bergstrom [...]
     To quote @pluralistic:
     "The problem isn’t that the chatbots lie all the time — it’s that they usually tell the truth, but then they start spouting confident lies."
     and:
     "That means that when Google ingests and repeats a lie, the lie gets spread to more sources. Those sources then form the basis for a new kind of ground truth, a “zombie statistic” that can’t be killed, despite its manifest wrongness."

     https://doctorow.medium.com/googles-ai-hype-circle-6158804d1299

   • Embed this notice
     DrYak (dryak@mstdn.science)'s status on Thursday, 18-May-2023 03:35:43 JST DrYak

     in reply to

     @ct_bergstrom Though this has still the usual drawbacks associated with using a LLM to generate an output for which the user doesn't have knowledge: there still risk that the Chat AI "lies"/hallucinates (i.e. Bullshits with confidence) by accident and the result is then taken for truth by the user.
     [...]

     Paul Cantrell repeated this.