Explore the implications of scaling up AI systems in this thought-provoking interview with Jacob Steinhardt. Delve into the concept of emergence in AI, examining how qualitative differences arise as systems scale up, using GPT-3 as a prime example. Investigate the intersection of engineering and philosophical perspectives on AI's future, and critically analyze the Paperclip Maximizer thought experiment in relation to AI safety and alignment. Examine potential failure modes in machine learning systems, including imitative deception and the generalization of empirical findings. Gain insights into current research on AI truthfulness and deception, and consider recommendations for improving AI alignment and safety as these systems continue to evolve and emerge in unexpected ways.
More Is Different for AI - Scaling Up, Emergence, and Paperclip Maximizers
Overview
Syllabus
Introduction
Start of Interview
Blog posts series
More Is Different for AI Blog Post
Do you think this emergence is mainly a property from the interaction of things?
How does phase transition or scaling-up play into AI and Machine Learning?
GPT-3 as an example of qualitative difference in scaling up
GPT-3 as an emergent phenomenon in context learning
Brief introduction of different viewpoints on the future of AI and its alignment
How does the phenomenon of emergence play into this game between the Engineering and the Philosophy viewpoint?
Paperclip Maximizer on AI safety and alignment
Thought Experiments
Imitative Deception
TruthfulQA: Measuring How Models Mimic Human Falsehoods Paper
ML Systems Will Have Weird Failure Models Blog Post
Is there any work to get a system to be deceptive?
Empirical Findings Generalize Surprisingly Far Blog Post
What would you recommend to guarantee better AI alignment or safety?
Remarks
Taught by
Yannic Kilcher
