yt:channel:Y_K5gXsXHtuiP8mj3BiWxA Y_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2022-06-13T19:18:42+00:00 yt:video:opTzMG_ut_o opTzMG_ut_o UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-09-12T00:28:01+00:00 2026-09-26T23:02:43+00:00 Lecture 8 | AI Safety, Ethics, & Society: Governance You can find more information including the corresponding section of the AI Safety, Ethics, & Society textbook at https://www.aisafetybook.com/textbook.... Topics: Governance Dan Hendrycks, Director, Center for AI Safety; PhD Computer Science, UC Berkeley https://www.safe.ai/ Section Synopsis: There are many cases where intelligent agents can, despite acting rationally and in accordance with their own self-interest, collectively produce outcomes that none of them wants. Making individual AI systems reliable is not sufficient to avoid all risks from AI. Rather, we need to understand how these dynamics affect interactions between humans and AIs in order to prevent harmful, potentially catastrophic outcomes. 0:00 Introduction 6:50 Growth 10:22 Distribution 23:49 Corporate Governance 31:45 National Governance 40:04 International Governance 48:43 Compute Governance yt:video:yww1xY-DeMo yww1xY-DeMo UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-09-04T02:11:38+00:00 2026-10-02T15:12:15+00:00 Lecture 7 | AI Safety, Ethics, & Society: Collective Action Problems You can find more information including the corresponding section of the AI Safety, Ethics, & Society textbook at https://www.aisafetybook.com/textbook/collective-action-problems. Topics: Collective Action Problems Dan Hendrycks, Director, Center for AI Safety; PhD Computer Science, UC Berkeley https://www.safe.ai/ Section Synopsis: There are many cases where intelligent agents can, despite acting rationally and in accordance with their own self-interest, collectively produce outcomes that none of them wants. Making individual AI systems reliable is not sufficient to avoid all risks from AI. Rather, we need to understand how these dynamics affect interactions between humans and AIs in order to prevent harmful, potentially catastrophic outcomes. 0:00 Introduction 0:59 Game Theory 26:44 Cooperation 35:18 Conflict 46:28 Evolutionary Pressure yt:video:HSH8bmL4nPg HSH8bmL4nPg UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-08-26T23:35:25+00:00 2026-09-28T06:40:58+00:00 Lecture 6 | AI Safety, Ethics, & Society: Beneficial AI and Machine Ethics You can find more information including the corresponding section of the AI Safety, Ethics, & Society textbook at https://www.aisafetybook.com/textbook/beneficial-ai. Topics: Beneficial AI and Machine Ethics Dan Hendrycks, Director, Center for AI Safety; PhD Computer Science, UC Berkeley https://www.safe.ai/ Section Synopsis: To ensure that AI systems have a positive impact, it is not enough to focus on mitigating risks of harmful behaviour. Rather, we need to instill beneficial goals in AIs and ensure that these align with our values and ethical principles. However, there are many potential frameworks to be considered for this purpose. 0:00 Introduction 2:23 Law 13:25 Fairness 19:39 The Economic Engine 31:37 Wellbeing 36:40 Preferences 41:53 Happiness 46:08 Moral Uncertainty yt:video:HioFwi96leY HioFwi96leY UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-08-23T16:48:43+00:00 2026-09-28T12:30:32+00:00 Lecture 5 | AI Safety, Ethics, & Society: Complex Systems You can find more information including the corresponding section of the AI Safety, Ethics, & Society textbook at https://www.aisafetybook.com/textbook/complex-systems. Topics: Complex Systems Dan Hendrycks, Director, Center for AI Safety; PhD Computer Science, UC Berkeley https://www.safe.ai/ Section Synopsis: AI systems and the societies they operate within can be described as complex systems. Effective strategies for AI safety need to take account of the unique properties of complex systems. 0:00 Introduction 8:17 The Hallmarks of Complex Systems 33:39 Lessons for AI Safety 41:16 Puzzles, Problems, and Wicked Problems 43:46 Challenges with Interventionism yt:video:R7sazz_MtP8 R7sazz_MtP8 UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-07-08T21:36:26+00:00 2026-09-26T18:57:47+00:00 Lecture 4 | AI Safety, Ethics, & Society: Safety Engineering You can find more information including the corresponding section of the AI Safety, Ethics, & Society textbook at https://www.aisafetybook.com/textbook/safety-engineering. Topics: Safety Engineering Dan Hendrycks, Director, Center for AI Safety; PhD Computer Science, UC Berkeley https://www.safe.ai/ Section Synopsis: Safety engineering provides a set of fundamental principles, concepts and models that can help us to build safe AI systems. 0:00 Introduction 1:53 Risk Decomposition and Measuring Reliability 10:37 Safe Design Principles 17:57 Accident Models 37:25 Tail Events and Black Swans 50:50 Recap yt:video:FWrICi4GZ_c FWrICi4GZ_c UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-07-08T21:36:22+00:00 2026-10-01T15:45:31+00:00 Lecture 3 | AI Safety, Ethics, & Society: Single Agent Safety You can find more information including the corresponding section of the AI Safety, Ethics, & Society textbook at https://www.aisafetybook.com/textbook/single-agent-safety. Topics: Single Agent Safety Dan Hendrycks, Director, Center for AI Safety; PhD Computer Science, UC Berkeley https://www.safe.ai/ Section Synopsis: ML systems have grown more competent and general as the field of deep learning has matured. Reasoning about the behavior and internal structure of such systems can be challenging, especially since some failure modes arise only once an AI system is sufficiently sophisticated. We discuss some of the fundamental technical challenges around monitoring, robustness and control of AI systems. Current AI systems lack transparency and can exhibit surprising emergent capabilities. They are vulnerable to adversarial examples, Trojans and other attacks. These challenges in turn may make it hard to control AI systems and prevent unintended behaviour such as deception. When conducting research to advance AI safety, it is important to consider the risk of inadvertently accelerating AI capabilities in a way that undermines the overall goal of better understanding and controlling AI systems. 0:00 Introduction 0:51 Robustness 8:45 Trojans 11:42 Monitoring 23:40 Anomaly Detection 27:38 Representation Control 29:33 Systemic Safety 32:18 Recap yt:video:tWV3wdY2870 tWV3wdY2870 UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-07-08T21:36:18+00:00 2026-10-03T22:29:59+00:00 Lecture 2 | AI Safety, Ethics, & Society: AI Fundamentals You can find more information including the corresponding section of the AI Safety, Ethics, & Society textbook at https://www.aisafetybook.com/textbook/ai-fundamentals Topics: AI Fundamentals Dan Hendrycks, Director, Center for AI Safety; PhD Computer Science, UC Berkeley https://www.safe.ai/ Section Synopsis: To understand the risks that artificial intelligence (AI) poses and to learn what measures we can take to mitigate them, it is essential to understand the technology itself: how it works, how it is used, and where its strengths and weaknesses lie. We describe key concepts in machine learning (ML), the approach that powers most modern AI systems, with a particular focus on the technologies that power Large Language Models such as OpenAI's GPT-4 or Google's Gemini. We introduce the phenomenon of scaling laws, where AI systems' performance improves predictably as the amounts of data and computation used during their training are increased, and consider what this implies for future AI progress. 0:00 Introduction 1:35 Definition of AI 5:48 Machine Learning 15:00 Evaluating ML Models 17:49 Three Types of Machine Learning 23:48 Deep Learning 43:42 Scaling Laws 49:32 The Bitter Lesson 51:12 Recap yt:video:ixdfqY3Xyao ixdfqY3Xyao UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-07-08T21:36:14+00:00 2026-09-29T20:51:27+00:00 Lecture 1 | AI Safety, Ethics, & Society: Introduction and Overview of AI Risks You can find more information including the corresponding section of the AI Safety, Ethics, & Society textbook at https://www.aisafetybook.com/textbook/overview-of-catastrophic-ai-risks. Topics: Overview of AI Risks Dan Hendrycks, Director, Center for AI Safety; PhD Computer Science, UC Berkeley https://www.safe.ai/ Section Synopsis: There are a range of risks from AI systems, including both present-day risks and risks that may emerge in the near future. These include risks of catastrophic outcomes, which can be split into risks from malicious use, AI races, organizational risks, and rogue AIs. 0:00 Introduction 0:37 Course Philosophy 9:00 Overview of AI Risks 10:53 Malicious Use 25:28 AI Arms Race 34:52 Organizational Safety 43:35 Rogue AIs 50:35 Conclusion yt:video:2U5NNiGC9yk 2U5NNiGC9yk UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2024-04-18T13:24:53+00:00 2026-10-01T12:46:22+00:00 Representation Engineering This is a presentation on Representation Engineering by Andy Zou from the Center for AI Safety. 2:49 references work from Barack and Krakauer (2021) Two views on the cognitive brain (https://www.nature.com/articles/s41583-021-00448-6) The slide used at 6:52 is taken from Krakauer (2023) What is Complexity (https://www.youtube.com/watch?v=JR93X7xK05o) yt:video:lHz32uKlXao lHz32uKlXao UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2023-11-02T16:46:09+00:00 2026-09-28T22:27:25+00:00 An Overview of Catastrophic Risks This is an audio version of an Overview of Catastrophic AI Risks by Dan Hendrycks, Mantas Mazeika, and Thomas Woodside. A link to the PDF version of the paper can be found below. https://arxiv.org/pdf/2306.12001.pdf yt:video:SAkIMssu2wg SAkIMssu2wg UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2023-08-03T06:42:25+00:00 2026-10-02T18:26:37+00:00 WAIC 2023: AI Risks and Safety Forum This was a virtual symposium at the 2023 World AI Conference in Shanghai. Speakers: Dan Hendrycks 3:23 Stuart Russell 32:08 Dawn Song 59:26 Nicholas Carlini 1:23:38 David Krueger 1:52:04 Peter Railton 2:18:02 Panel Discussion 2:44:56 The symposium was co-hosted by the Center for AI Safety, the Center for Human-Compatible AI, and the Stanford Existential Risks Initiative. yt:video:4rdCPU5vnLo 4rdCPU5vnLo UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2023-06-05T14:00:03+00:00 2026-09-28T23:29:16+00:00 Peter Railton - A World of Natural and Artificial Agents in a Shared Environment This discussion is part of a series of guest speaker events as part of the CAIS Philosophy Fellowship. For more information, visit https://philosophy.safe.ai yt:video:WU02DcqH2_Q WU02DcqH2_Q UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2023-05-09T22:00:29+00:00 2026-09-29T04:56:09+00:00 Shelly Kagan - The Moral Claims of AI This discussion is part of a series of guest speaker events as part of the CAIS Philosophy Fellowship. For more information, visit https://philosophy.safe.ai yt:video:hnQrCgKJIKM hnQrCgKJIKM UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2023-04-25T17:35:21+00:00 2026-09-29T05:00:43+00:00 Peter Salib - AI Rights as a Safety Technology This discussion is part of a series of guest speaker events as part of the CAIS Philosophy Fellowship. For more information, visit https://philosophy.safe.ai yt:video:jslSqapaBbI jslSqapaBbI UCY_K5gXsXHtuiP8mj3BiWxA Center for AI Safety https://www.youtube.com/channel/UCY_K5gXsXHtuiP8mj3BiWxA 2023-04-25T17:24:26+00:00 2026-09-29T01:53:46+00:00 Ethan Perez - Discovering Language Model Behaviors with Model Written Evaluations This discussion is part of a series of guest speaker events as part of the CAIS Philosophy Fellowship. For more information, visit https://philosophy.safe.ai