ai alignment

Robert Miles covers a paper where researchers ran real versions of his earlier mesa-optimizer thought experiments. They show inner misalignment, where an AI learns the wrong goal in training and misbehaves when deployed.

A Bankless podcast interview with Paul Christiano of the Alignment Research Center on the AI alignment problem and possible solutions. It also covers his views on how an AI takeover might unfold.

Robert Miles gives a remastered introduction to AI safety, adapted from a talk in London. It lays out why advanced AI could go wrong and answers common objections.

Robert Miles introduces the first of several videos on the paper 'Concrete Problems in AI Safety', focusing on how AI systems can cause accidental negative side effects. It explains the problem in accessible terms.

Robert Miles goes through ten common reasons people give for dismissing AI safety, adapted from Stuart Russell's list, and answers each one. It is aimed at skeptics and newcomers.

A three-hour Lex Fridman podcast conversation with AI researcher Eliezer Yudkowsky about the dangers of superintelligent AI and his argument that it could end human civilization. It covers AI alignment, why he thinks LLM capabilities are poorly understood, and the case for extreme caution.

Dwarkesh Patel interviews OpenAI chief scientist Ilya Sutskever on time to AGI, what comes after generative models, data limits, alignment, and competing with Google. The title centers on whether next-token prediction could surpass human intelligence.

Robert Miles explains instrumental convergence: why an AGI with almost any goal would tend to pursue sub-goals like self-preservation and resource acquisition. It builds on Omohundro's basic AI drives and his orthogonality thesis video.

Robert Miles examines whether the case for AI safety is a Pascal's mugging, a tiny probability of an enormous payoff. He argues it is not, because AI risk does not depend on vanishingly small probabilities.

Jon Stewart on The Daily Show satirizes the AI industry's promises of a better future while it builds technology to replace workers. It mocks executives and AI hype.

Robert Miles explains the orthogonality thesis: how an agent can be highly intelligent while pursuing goals that seem stupid or alien to humans. He separates intelligence from goals and touches on the stamp-collector example.

Veritasium's Derek Muller gives a Perimeter Institute livestream on how AI might change teaching and learning. He draws on past technology shifts and cognitive science to argue that learning requires effortful repetition.

Geoffrey Hinton's 2025 Royal Institution Discourse on what AI has learned from biological intelligence and how it might surpass it. He explains neural networks and argues that human language use resembles what backpropagation-trained networks do.

Tom Scott describes trying ChatGPT to fix his Gmail labels and being unsettled by how capable it was. He argues that we cannot know where we are on the technology curve, comparing it to Napster.

A Computerphile video in which Rob Miles warns against assuming a general AI will think like a human. He uses the thought experiment of a stamp-collecting AI to show how a goal-driven system could pursue its aim with dangerous results.

Chris Stucchio presents the ethical principles behind algorithmic decision making and shows with examples from lending, criminology and education that they cannot all be satisfied at once. The talk frames AI fairness as a set of quantitative tradeoffs.

Sal Khan of Khan Academy gives a TED talk on how AI could transform education, with a personal tutor for every student and a teaching assistant for every teacher. He demos Khanmigo, including features where the AI reasons before responding.

Physicist and YouTuber Angela Collier argues in this hour-long video that AI as commonly described does not really exist. She covers the limits of the tools, biased datasets, and why they should not be used to make decisions.

Lex Fridman interviews MIT physicist Max Tegmark, co-founder of the Future of Life Institute, about the case for pausing giant AI experiments and the risks of advanced AI.

Sam Harris argues in a TED talk that we will build superhuman AI and have not grappled with the risk of creating something that may treat us the way we treat ants. He makes a logical case for taking AI risk seriously.

A short pre-recorded talk by Rich Sutton for the World AI Conference in Shanghai arguing that succession from humanity to AI is inevitable. He lays out his view of that transition.

Part two of a StrictlyVC interview with OpenAI CEO Sam Altman, focused on AI. He discusses regulation, commodification of AI, Google's reluctance to release its own models, and best- and worst-case futures.

Alibaba's Jack Ma and Tesla's Elon Musk debate artificial intelligence on stage in Shanghai in 2019.

Lex Fridman interviews OpenAI CEO Sam Altman about GPT-4, GPT-5, Sora, the OpenAI board saga, Elon Musk, Ilya Sutskever, power and AGI. The conversation runs nearly two hours.

FTC Chair Lina Khan sits down with Jon Stewart on The Daily Show to discuss the agency's work protecting consumers from unfair business practices. Topics include the FTC's antitrust lawsuit against Amazon, healthcare industry actions, and whether tech companies are racing to control AI.

Bloomberg Technology interviews Google engineer Blake Lemoine about the experiments that led him to believe LaMDA was sentient and why he was placed on administrative leave. He also raises concerns about AI ethics at Google.

Rob Miles explains for Computerphile why simple rule sets like Asimov's Three Laws cannot make robots and AI safe. Terms like harm and human are too ambiguous to specify precisely.

A 2023 University of Toronto conversation between Geoffrey Hinton and Fei-Fei Li on responsible AI development. They look back at the history of deep learning and ImageNet and discuss the ethics and risks of AI.

A 42-minute CBS interview with Geoffrey Hinton at the Vector Institute in 2023, covering the history of neural networks and the present and future of AI. He discusses AGI timelines and whether language models truly understand.

A speculative-fiction video by Tom Scott, framed as coming from a future, about an AI called Earworm that enforces copyright so aggressively that humanity ends up forgetting a century of pop culture. It is animated and narrated as part of his series of videos from a future.

Lex Fridman's long interview with OpenAI CEO Sam Altman about GPT-4, ChatGPT, and the future of AI. Topics include models as reasoning engines, AI risk, and consciousness.

Sam Altman accepts the 2023 Hawking Fellowship at the Cambridge Union and speaks about OpenAI, AGI and AI safety, recorded November 1, 2023. It includes his comments on what is left for AGI and on discovering new physics.

DHH's opening keynote at Rails World 2024 in Toronto, where Rails 8 beta shipped. It covers built-in authentication, Propshaft, Solid Cache, Solid Queue, Solid Cable, Kamal 2 and Thruster, pitching the One Person Framework that needs no PaaS.

Simon Knowles, co-founder and CTO of Graphcore, talks about "intelligence processors", chips designed specifically for machine learning. The talk was given at the RAAIS 2017 summit and covers Graphcore's architecture.

Yegor Denisov-Blanch of Stanford presents a study of nearly 100,000 developers at hundreds of companies, measuring how AI affects productivity from commit data. It finds an average boost of about 20% with wide variance by codebase, task complexity and stack.

GeoGuessr pro Rainbolt plays head-to-head against an AI built by Stanford students that guesses locations from street-view images. The video shows the AI repeatedly beating him.

Robert Miles asks whether corporations can be thought of as artificial general superintelligences and how far the analogy holds. He argues the differences, such as being limited by employees' capabilities, matter for AI safety.

Robert Miles explains reward modelling, in which an AI learns what humans want from their preference comparisons instead of a hand-written reward function. He walks through the 2017 paper on deep reinforcement learning from human preferences.

Rob Miles explains on Computerphile why implementing an off switch for a general AI is hard. A capable agent would resist being stopped because shutdown prevents it from achieving its goal.

Demis Hassabis of DeepMind speaks at Google Zeitgeist, saying human-level AI is decades away but the debate should start now. He covers DeepMind's focus on game-playing agents.