The OTHER AI Alignment Problem: Mesa-Optimizers and Inner Alignment Published 2021-02-16 Download video MP4 360p Recommendations 11:47 We Were Right! Real Inner Misalignment 17:52 Training AI Without Writing A Reward Function, with Reward Modelling 11:32 There's No Rule That Says We'll Make It 17:38 The moment we stopped understanding AI [AlexNet] 15:38 A Response to Steven Pinker on AI 19:42 Space-Time: The Biggest Problem in Physics 20:00 AI "Stop Button" Problem - Computerphile 16:29 10 Reasons to Ignore AI Safety 13:41 Is AI Safety a Pascal's Mugging? 09:24 Why Does AI Lie, and What Can We Do About It? 27:48 AI & Logical Induction - Computerphile 10:41 What can AGI do? I/O and Speed 22:38 How Will We Know When AI is Conscious? 16:20 The REAL Three Body Problem in Physics 07:23 AI Safety Gridworlds 14:28 Can we build AI without losing control over it? | Sam Harris 16:00 AI Safety Gym - Computerphile 07:42 AI? Just Sandbox it... - Computerphile 13:54 The True Story of How GPT-2 Became Maximally Lewd Similar videos 01:39 What happens if AI alignment goes wrong, explained by Gilfoyle of Silicon valley. 1:09:33 Evan Hubinger - The Inner Alignment Problem 10:20 Deceptive Misaligned Mesa-Optimisers? It's More Likely Than You Think... 1:37:06 Evan Hubinger on Inner Alignment, Outer Alignment, and Proposals for Building Safe Advanced AI 03:00 AI Alignment Problem: Extremes of Optimization 1:13:48 The Alignment Problem: Machine Learning and Human Values with Brian Christian 03:48 Why is AI Alignment So Difficult? 12:35 Buddhism and the AI "Alignment Problem" 58:26 The Alignment Problem - Brian Christian 02:22 Mesa Optimization Algorithms in Transformers More results