InAI AlignmentbyPaul Christiano·Apr 27, 2023My views on “doom”I’m often asked: “what’s the probability of a really bad outcome from AI?” In this post I answer 10 versions of that question.A response icon3A response icon3
InAI AlignmentbyPaul Christiano·Dec 27, 2022Can we efficiently distinguish different mechanisms?Can a model produce coherent predictions based on two very different mechanisms without there being any efficient way to distinguish them?
InAI AlignmentbyPaul Christiano·Dec 16, 2022Can we efficiently explain model behaviors?It may be impossible to automatically find explanations. That would complicate ARC’s alignment plan, but our work can still be useful.
InAI AlignmentbyPaul Christiano·Dec 13, 2022AI alignment is distinct from its near-term applicationsNot everyone will agree about how AI systems should behave, but no one wants AI to kill everyone.A response icon1A response icon1
InAI AlignmentbyPaul Christiano·Dec 1, 2022Finding gliders in the game of lifeWalking through a simple concrete example of ARC’s approach to ELK based on mechanistic anomaly detection.
InAI AlignmentbyPaul Christiano·Nov 25, 2022Mechanistic anomaly detection and ELKAn approach to ELK based on finding the “normal reason” for model behaviors on the training distribution and flagging anomalous exaples.
InAI AlignmentbyPaul Christiano·Feb 25, 2022Eliciting latent knowledgeHow can we train an AI to honestly tell us when our eyes deceive us?A response icon1A response icon1
InAI AlignmentbyPaul Christiano·Jun 13, 2021Answering questions honestly given world-model mismatchesI expect AIs and humans to think about the world differently. Does that make it more complicated for an AI to “honestly answer questions”?
InAI AlignmentbyPaul Christiano·Jun 10, 2021A naive alignment strategy and optimism about generalizationI describe a very naive strategy for training a model to “tell us what it knows.”
InAI AlignmentbyPaul Christiano·May 28, 2021Teaching ML to answer questions honestly instead of predicting human answersI discuss a three step plan for learning to answer questions honestly instead of predicting what a human would say.
