Replies

• 

  Chris 👾chris@mstdn.games

  May 24, 2025, 9:21 AM

  @minimalparts It's a ridiculous made-up PR story and the tech media should be more sceptical and thoroughly instead of just jumping on the narrative of a "human-like", capable, evil AI for clicks.

  In this specific test, the model literally had only two options: " In order to elicit this extreme blackmail behavior, the scenario was designed to allow the model no other options to increase its odds of survival; the model’s only options were blackmail or accepting its replacement."

  #Anthropic

  💬 1🔄 0⭐ 0
• 

  Aurelie Herbelot (she/her)minimalparts@denotation.link

  May 24, 2025, 9:39 AM

  @chris Totally! It also shows how model evaluation has totally gone out of the window. How can you even call it 'a test' when all you did was prompt the model a couple of times on one specific scenario and see what happens? But as you say, it's all click-bait.

  💬 0🔄 0⭐ 0