Can AI cause the end of the human race
summary
Covers specific milestones and risks: chess, Go and Atari agents learn by reinforcement learning through tens or hundreds of thousands of self-play episodes; AlphaGo Zero reaches master level in days; hardware improves cost-per-FLOP (Nvidia Volta claims ~10x over prior generation); Google reports 40% cooling reduction and now lets AI control datacenter cooling. Lists concrete harms: mass job displacement, autonomous mini-drones with grams of explosive, targeted spear-phishing and deepfakes, and an AI alignment failure risk experts put around 50% by 2040–2050.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:58First things first
- 1:48Current state
- 3:46RL is the way to go in Al
- 8:32After a problem is solved
- 9:41120 Years of "Moore's Law"
- 11:33Predictions about AI
- 13:23Financial incentives
- 14:38Where does the tallent go
- 15:46The good stuff
- 16:20The Bad Stuff
- 20:31Ban Lethal Al Weapons
- 21:20Malicious uses of AI
- 24:44Al Alignment
- 26:03Don't panic!
- 26:59But do help if you can
related talks
21mWhen the machines out-smarts us
19mDealing with AI as a Dystopian Threat to Humanity
48mThe Perils and Promise of Artificial Intelligence
19mUpload or die: apocalyptic visions in AI
30mIntroduction to AI Safety. Let's be ready before it's too late!
23mThe AI Dilemma — Steering towards a More Humane Future
54mAI vs Humans: Rewriting the Rules of Success
05mAI is your children, what is our future with them?