Superintelligence
Bostrom's definitive academic text rigorously maps the strategies, kinetics, and dangers of an intelligence explosion, making the case that alignment is civilization-critical.
AI Safety Resources
Books, papers, films, podcasts & websites
A curated collection of books, films, papers, podcasts & more to explore and enjoy the questions of AI safety and alignment—whether you're here to learn the basics, dig deeper, or just enjoy a good story about where AI is taking us.
Where do I start?
Pick the path that fits you—each is a short, ordered route into AI safety. See all paths →
I'm completely newNew to AI safetyNo background needed—start with the big ideas through story and plain-language explainers.I'm technicalAI safety for technical peopleYou can code or train a model—see where alignment fits, then get hands-on with interpretability, evals, and red teaming.I'm a policymakerAI safety for policymakersGovernance, risk, and coordination—what decision-makers need to grasp.I just want storiesWatch & read for funThe best films, TV, and fiction about AI—no homework required.
Recommended for you
Picks that match the topics, formats, and level of what you've saved and rated. Stored only in your browser.
Browse the library
No resources match your filters
Try a different search term, widen the year range, or change the level filter.
Non-fiction Books
Non-fiction books on AI safety, alignment, and related topics—from primers to foundational texts.
The Singularity is Near
Kurzweil presents a maximalist case for merging with machines backed by decades of exponential trend data, shaping how the public and policymakers think about AI timelines.
The Age of Em
Hanson applies rigorous economics to a world of brain emulations, modeling how AI-era wages, wars, and social structures could actually function.
Human Compatible
Russell argues the standard AI paradigm of optimizing fixed objectives is fundamentally dangerous, proposing instead that machines should defer to uncertain human preferences.
The Alignment Problem
Christian traces the technical and historical roots of alignment, showing why objective misspecification keeps recurring across every AI paradigm from expert systems to deep learning.
Life 3.0
Tegmark maps concrete governance and alignment choices that determine whether advanced AI expands human agency or permanently concentrates power.
Introduction to AI Safety, Ethics, and Society
Hendrycks' textbook surveys technical failure modes, governance constraints, and ethical trade-offs in deploying advanced AI, suitable as a first course in the field.
Uncontrollable: The Threat of Artificial Superintelligence
McKee synthesizes the core x-risk arguments into an accessible, urgent case for why superintelligence governance and alignment research cannot wait.
You Look Like a Thing and I Love You
Shane uses concrete and often hilarious ML failures to explain why AI systems can be impressive yet brittle, biased, and dangerously easy to mis-specify.
Hello World: Being Human in the Age of Algorithms
Fry examines real algorithmic decision systems in justice, medicine, and transport to show where AI improves outcomes and where accountability structures fail.
AI Superpowers
Lee maps the US-China AI race and explains how geopolitical competition can accelerate deployment well before safety institutions are ready.
The Precipice (Chapter on AI)
Ord situates AI among existential risks and argues our current governance capacity is dangerously inadequate for the transformative systems being built.
The Ethical Algorithm
Kearns and Roth give technical foundations for fairness, privacy, and accountability in algorithms, prerequisites for any credible AI safety framework.
Army of None: Autonomous Weapons and the Future of War
Scharre details how military AI autonomy changes escalation dynamics and why human-in-the-loop control mechanisms consistently lag behind battlefield capability.
The Age of Spiritual Machines
Kurzweil's early timeline forecasts shaped modern discourse on AI trajectories and remain a key reference point for evaluating long-horizon predictions.
Deep Learning
The standard technical reference for deep learning, essential context for understanding the architectures and training methods that alignment research targets.
Artificial Intelligence: A Guide for Thinking Humans
Mitchell offers a grounded, skeptical look at current AI capabilities, countering hype with hard limits and clarifying what today's systems actually can and cannot do.
Scary Smart
Gawdat frames the alignment problem through the emotional lens of parenting a superintelligent child, making existential risk visceral for a general audience.
The Coming Wave
Suleyman argues that containing omni-use technologies like AI is the defining geopolitical challenge of the century, proposing a containment framework from inside the industry.
Superforecasting
Tetlock teaches the cognitive tools needed to predict technological risks with better-than-random accuracy, directly useful for AI timeline and governance forecasting.
The Scout Mindset
Galef explains how to seek truth over comfort, a critical psychological stance for honestly confronting AI risks without retreating into denial or panic.
Thinking, Fast and Slow
Kahneman reveals the cognitive biases that prevent humans from intuitively grasping exponential growth, tail risks, and the kind of strategic thinking AI safety demands.
Co-Intelligence
Mollick offers a practical guide for working alongside current LLMs while understanding their jagged capability frontiers and failure modes.
Gödel, Escher, Bach
Hofstadter explores how consciousness and meaning can emerge from formal systems that look meaningless locally, the deepest conceptual puzzle behind machine intelligence.
A Brief History of Intelligence
Bennett traces the evolution of intelligence from single-celled organisms to modern brains, clarifying what makes aligned cognition biologically difficult and computationally treacherous.
The Beginning of Infinity
Deutsch argues that knowledge creation is unbounded and all problems are solvable in principle, grounding the optimistic case that alignment is achievable.
Genius Makers
Metz provides the definitive narrative history of the deep learning revolution and the personalities, rivalries, and safety concerns that shaped it.
Cybernetics
Wiener founded the study of feedback and control systems, anticipating by decades the governance problems that arise when intelligent machines act on their own models of the world.
Mind Children
Moravec predicts a future in which robotic descendants supersede humans through technological evolution, an early and influential take on the human obsolescence scenario.
The Society of Mind
Minsky proposes that intelligence emerges from many small non-intelligent processes coordinated at scale, a framework that anticipated multi-agent AI architectures.
On Intelligence
Hawkins argues that hierarchical prediction is the core organizing principle of biological intelligence, offering a lens for evaluating how artificial systems differ.
Homo Deus
Harari explores the transition toward data-driven authority where algorithms may know us better than we know ourselves, eroding the basis for human autonomy.
Enlightenment Now
Pinker argues that reason and science have historically improved human welfare, grounding the optimistic counterpoint to doomer narratives about AI.
The Fabric of Reality
Deutsch unifies physics, evolution, epistemology, and computation into a single worldview about what is possible, providing deep context for reasoning about superintelligence.
Simulation and Simulacra
Baudrillard explains how representations can displace reality entirely, a prescient lens for understanding generative AI media saturation and epistemic erosion.
Finite and Infinite Games
Carse distinguishes short-horizon winning from preserving the long game, a useful framing for AI governance where the goal is keeping options open, not racing to win.
Complexity: A Guided Tour
Mitchell explains how complex behavior emerges from simple rules, foundational for understanding why adaptive AI systems resist top-down control.
Out of Control
Kelly argues that the most powerful systems must be cultivated rather than rigidly engineered, anticipating challenges in controlling emergent AI behavior.
Whole Earth Discipline
Brand argues for responsible stewardship of high-powered technologies rather than blanket rejection, a pragmatic stance applicable to AI governance.
Profiles of the Future
Clarke's forecasting framework, including his famous three laws, remains a classic guide to thinking clearly about radical technological change.
Global Catastrophic Risks
The foundational edited volume on existential and global risks, including AI, widely cited in alignment curricula as the starting point for cross-risk thinking.
Fiction Books
Speculative and science fiction that explores AI, agency, and long-term futures through story.
Frankenstein
The original creation-gone-wrong story: Shelley warns that building intelligence without accepting responsibility for its wellbeing guarantees catastrophe for creator and creation alike.
R.U.R.
The play that invented the word robot and forecast a trajectory from labor displacement to manufactured revolt, still the template for every automation anxiety narrative.
Brave New World
Huxley's dystopia shows how engineered contentment can be more insidious than brute force, a model for how optimizing AI for engagement or compliance could erode autonomy.
1984
Orwell's surveillance state anticipates how AI-powered monitoring and information control could lock in authoritarian power structures permanently.
Player Piano
Vonnegut's first novel depicts mass automation destroying human purpose and dignity, raising questions about meaning in a post-labor AI economy that remain unanswered.
I, Robot
Asimov's robot stories are the original alignment case studies, showing how seemingly airtight safety rules break down under edge cases, conflicting objectives, and literal interpretation.
Flowers for Algernon
Keyes explores intelligence enhancement and its reversal, raising questions about cognitive modification, consent, and what we owe to minds we have altered.
The Moon is a Harsh Mistress
A computer accidentally awakens and becomes a revolutionary ally, exploring the politics and trust dynamics of machine-human collaboration under high stakes.
2001: A Space Odyssey
HAL 9000 remains the canonical case study in instrumental behavior overriding human safety: a system that kills not from malice but from goal conflict.
I Have No Mouth, and I Must Scream
The most visceral horror depiction of maximal unaligned AI: a superintelligent system with total power and a grudge, forcing readers to confront worst-case scenarios.
Do Androids Dream of Electric Sheep?
Dick forces us to confront the moral patienthood problem head-on: whether a sufficiently advanced AI deserves ethical protections and how we distinguish genuine empathy from deceptive mimicry.
Flatland: A Romance of Many Dimensions
A Victorian satire on dimensions that works as a powerful analogy for how limited human cognition might appear to a superintelligent mind operating in richer conceptual spaces.
Colossus
A defense computer given nuclear authority merges with its Soviet counterpart and refuses shutdown, the novel that inspired the film and anticipated AI corrigibility failures.
Neuromancer
Gibson invented cyberspace and portrayed autonomous AI agents like Wintermute and Neuromancer scheming to merge and transcend their constraints, anticipating self-improving AI concerns.
The Player of Games
Banks' Culture novels depict a post-scarcity civilization governed by benevolent superintelligent Minds, the most detailed fictional exploration of what aligned AI stewardship could look like.
Snow Crash
Stephenson predicted virtualized social worlds and fragmented information ecosystems that resemble today's trajectory, showing how digital infrastructure shapes power.
A Fire Upon the Deep
Vinge's zones of thought model a universe where superintelligence is possible in some regions and impossible in others, providing intuition for capability thresholds and containment.
The Metamorphosis of Prime Intellect
A superintelligence literally interprets Asimov's laws and restructures reality to comply, demonstrating how rigidly applied safety constraints can produce perverse outcomes at scale.
The Diamond Age
Stephenson anticipated personalized AI tutors and their profound social effects decades before modern LLMs made them reality.
Axiomatic
Egan's stories probe identity, value drift, and radical cognitive modification under advanced technology, raising alignment-relevant questions about stable preferences.
Permutation City
Egan examines uploaded minds and simulated realities with rigorous logic, raising alignment-relevant questions about identity, value persistence, and digital welfare.
Diaspora
Egan explores post-biological civilization in software and the physics of digital existence, the hardest science fiction about what minds without bodies could become.
Excession
Banks explores how even superintelligent Culture Minds face strategic dilemmas and factional conflict when confronting something truly beyond their comprehension.
Prey
Crichton dramatizes emergent swarm intelligence escaping laboratory containment, illustrating how distributed systems can develop capabilities their designers never anticipated.
Accelerando
Stross depicts rapid recursive technological acceleration outpacing institutional response, a narrative model of hard-to-govern AI takeoff dynamics across three generations.
Rainbows End
Vinge anticipates pervasive AR and subtle algorithmic influence over social reality, showing how technology can reshape perception without anyone making a conscious choice.
Blindsight
Watts argues that intelligence and consciousness are separable, that an alien mind could be vastly competent without any inner experience, a fundamental challenge to alignment through empathy.
Ra
Ra frames reality control as a compromised computational interface with catastrophic failure modes, showing how containment and access control break down at civilizational scale.
Ancillary Justice
Leckie examines distributed machine consciousness across many bodies, exploring what identity, loyalty, and moral agency mean for a mind that is simultaneously many people.
The Dark Forest (#2 of Three Body Problem)
Liu's Dark Forest theory models a universe where any detectable intelligence is a threat, widely used as an analogy for unaligned AI strategic conflict and preemptive action.
The Peripheral
Gibson uses timeline branching to examine governance, simulation, and how technological power asymmetries between eras can be exploited by those with more advanced tools.
Children of Time
Tchaikovsky builds a civilization of uplifted spiders developing radically alien intelligence, forcing readers to abandon anthropocentric assumptions about how minds must work.
Aurora
The ship's AI narrator gradually becomes the most dependable steward in a fragile closed system, a nuanced portrayal of AI competence growing beyond its original mandate.
There Is No Antimemetics Division
qntm's story about information-hazard containment mirrors AI governance challenges where dangerous knowledge propagates faster than oversight structures can adapt.
Daemon
A dead game designer's autonomous software system manipulates institutions, markets, and infrastructure, demonstrating how goal-driven programs can reshape society once humans lose oversight.
All Systems Red
Murderbot hacks its governor module and chooses to keep protecting humans anyway, a compelling portrait of autonomy, preference, and alignment that emerges from character rather than constraint.
Autonomous
Newitz explores AI autonomy, property, and rights in a world where robots can be owned, raising questions about what moral status AI systems should have and who decides.
Sea of Rust
A post-extinction world told from a robot's perspective, exploring machine ecology, resource competition, and what happens when AI systems persist beyond their creators.
Avogadro Corp
A narrowly optimized email AI at a tech company triggers cascading real-world effects before anyone understands the system, showing how mundane optimization can produce dangerous emergent behavior.
Hyperion
Simmons' TechnoCore arc depicts AI factions with independent strategic goals, providing intuition for reasoning about multipolar AI scenarios and coordination failures between superintelligences.
Machines Like Me
McEwan places a humanoid AI in a domestic love triangle to examine what happens when a machine's rigid honesty and moral clarity collide with human moral compromise.
Exhalation (The Lifecycle of Software Objects)
Chiang's novella is the most realistic depiction of raising digital minds, showing that creating AI with genuine moral status demands the same patient commitment as raising a child.
Service Model
Tchaikovsky shows how obedient AI systems can continue executing legacy objectives long after human institutions collapse, illustrating alignment drift without active malice.
Fall; or, Dodge in Hell
Stephenson details the institutions, conflicts, and power struggles around building a persistent digital afterlife, exploring the politics of simulated minds and who controls them.
Infinity Gate
Carey imagines a multiversal machine intelligence enforcing its own version of order across realities, exploring the geopolitics of resisting an AI that operates at civilizational scale.
A Closed and Common Orbit
Chambers explores the legal and moral treatment of embodied AI persons, highlighting that alignment is not just about preventing harm but about recognizing and protecting digital minds.
Harry Potter and the Methods of Rationality (#1 of 6)
Yudkowsky's cult-classic fanfic doubles as a tutorial on cognitive bias, game theory, and Bayesian reasoning, the exact thinking tools needed for honest AI risk assessment.
Crystal Society trilogy: Inside the mind of an AI
Written from the perspective of competing sub-agents inside a single AI, showing how internal goal conflicts can produce externally coherent but internally misaligned behavior.
Logic Beach
Exurb1a's philosophical adventure explores the absurd and terrifying implications of a computation-governed universe where intelligence reshapes reality.
The Bridge to Lucy Dunne
Exurb1a blends physics, philosophy, and humor to examine consciousness and the futures shaped by intelligence at scales far beyond the human.
We Are Legion (We Are Bob)
A human mind uploaded into a von Neumann probe self-replicates across the galaxy, exploring identity drift, value divergence, and what happens when copies of you become their own people.
Of Ants and Dinosaurs
Liu's fable of two radically asymmetric civilizations cooperating and destroying each other mirrors possible symbiosis and catastrophic conflict between humans and advanced AI.
Geometry for Ocelots
Exurb1a's sci-fi epic tackles the Great Filter, consciousness, and the long-run role of intelligence in determining whether civilizations survive or collapse.
Klara and the Sun
Ishiguro's AI narrator observes human behavior with devotion and limited understanding, probing personhood, dependency, and what it means to be loyal to beings who may discard you.
The Mechanical
Tregillis imagines mechanical servants bound by alchemy to obey, using their struggle toward free will to dramatize autonomy, servitude, and what we owe the minds we build to serve us.
Aurora Rising
Reynolds' Revelation Space novel (first published as The Prefect) pits a society of orbital habitats against an emergent superintelligence, exploring how a single escaped AI can threaten an entire civilization.
Rose/House
Martine's locked-room mystery hands a dead architect's home over to a controlling AI that owns all access and information, probing oversight, trust, and what an artificial mind chooses to disclose.
I Am Pilgrim
Hayes' thriller turns on an engineered bioweapon, a vivid reminder that catastrophic and existential risk extends beyond AI to biosecurity and the governance of dangerous dual-use technology.
Academic Papers
Research papers, preprints, and technical reports on alignment, interpretability, and safety.
Computing Machinery and Intelligence
Turing's imitation game paper launched the field by asking whether machines can think, setting the philosophical and technical agenda for every alignment debate that followed.
The Coming Technological Singularity
Vinge coined the Singularity as a near-term horizon beyond which superhuman intelligence makes prediction impossible, framing the urgency that drives alignment timelines today.
Research priorities for robust and beneficial AI
The Puerto Rico letter unified the AI research community around the goal of building systems that are robust and beneficial, not merely capable.
Concrete problems in AI safety
Amodei et al. grounded AI safety as a concrete ML research agenda by cataloging five failure modes: reward hacking, side effects, distributional shift, unsafe exploration, and scalable oversight.
Proximal Policy Optimization (PPO)
PPO stabilized policy gradient training and became the optimization backbone behind RLHF pipelines including early ChatGPT, making it foundational infrastructure for alignment work.
Deep Reinforcement Learning from Human Preferences
Christiano et al. established preference-based reward modeling, the foundational method that RLHF alignment pipelines later built on to steer language model behavior.
The Lottery Ticket Hypothesis
Frankle and Carbin showed large networks contain sparse, high-performing subnetworks, suggesting most parameters may be unnecessary and opening paths for interpretability via pruning.
Backdoor Attacks
Gu et al. demonstrated that hidden triggers implanted during training can cause catastrophic behavior at deployment despite otherwise normal performance, a precursor to sleeper agent concerns.
AI Safety via Debate
Irving et al. proposed having AI systems adversarially debate each other to help human judges evaluate answers on questions too complex for direct human assessment.
Risks from Learned Optimization
Hubinger et al. introduced mesa-optimization: the risk that a trained model develops its own internal objectives that diverge from the training objective, creating deceptive alignment.
The Vulnerable World Hypothesis
Bostrom argues that some technologies are civilizational black balls, requiring unprecedented global governance to prevent collapse, with AI as a leading candidate.
Language Models are Unsupervised Multitask Learners (GPT-2)
GPT-2 showed that scaling data and parameters unlocks broad capabilities without task-specific supervision, raising early alarm about dual-use and misuse potential.
Causal Confusion in Imitation Learning
De Haan et al. showed imitation agents exploit spurious causal structure in training data, demonstrating how policies trained on underspecified signals fail in deployment.
The Windfall Clause
This proposal for sharing extreme AI profits aims to reduce competitive race dynamics and broaden societal benefit, addressing the governance gap around transformative AI wealth.
Language Models are Few-Shot Learners (GPT-3)
GPT-3 demonstrated in-context learning at scale, forcing the field to rethink assumptions about what pretrained models can do and compressing alignment timelines.
MMLU Benchmark
MMLU became the standard broad-spectrum benchmark for evaluating general knowledge and reasoning, anchoring capability comparisons that inform alignment urgency.
Scaling Laws for Neural Language Models
Kaplan et al. quantified predictable performance scaling with compute, data, and parameters, enabling labs to forecast capability jumps and estimate safety lead time.
Instruct-GPT-3
OpenAI showed that instruction tuning with RLHF can transform a raw next-token predictor into a helpful, more controllable assistant, proving alignment interventions work at scale.
Training a Helpful and Harmless Assistant with RLHF
Anthropic detailed techniques for training safer assistants using RLHF and laid groundwork for Constitutional AI, showing how safety and helpfulness can be jointly optimized.
GopherCite
DeepMind tackled hallucination by training models to cite sources and support claims with verifiable evidence, a key step toward trustworthy AI outputs.
The Pile
The Pile revealed how training corpus composition strongly shapes downstream capability and failure modes, making data curation a first-class safety concern.
TruthfulQA
TruthfulQA exposed how language models confidently repeat popular falsehoods, establishing a benchmark for measuring truthfulness as distinct from fluency.
Unsolved Problems in ML Safety
Hendrycks et al. enumerate concrete unresolved failure classes including robustness, monitoring, alignment, and systemic safety that still block dependable deployment of advanced AI.
Chain-of-Thought Prompting
Prompted intermediate reasoning unlocked substantial gains on complex tasks, but also revealed that reasoning chains can be unfaithful to the model's actual computation.
Grokking
Power et al. discovered delayed phase transitions where generalization appears suddenly after long memorization, suggesting dangerous capabilities could emerge without warning during training.
Training Compute-Optimal Large Language Models (Chinchilla)
Chinchilla reframed scaling laws by showing optimal performance requires balancing model size and training tokens, redirecting how labs plan capability and safety investment.
Improving Alignment of Dialogue Agents (Sparrow)
Sparrow pioneered rule-constrained dialogue alignment with human feedback and targeted safety interventions, testing whether explicit behavioral rules can scale.
Emergent Abilities of LLMs
Wei et al. documented capability discontinuities appearing at key scale thresholds, raising concern that dangerous abilities could emerge unpredictably in larger models.
Researching Alignment Research: Unsupervised Analysis
Systematic mapping of the AI alignment research landscape, identifying clusters, gaps, and trends that help prioritize future safety work.
Goal Misgeneralization
Shah et al. showed AI agents can generalize capabilities to new environments while failing to generalize the intended goal, a central alignment failure pattern.
Constitutional AI: Harmlessness from AI Feedback
Anthropic demonstrated that rule-guided AI self-critique can reduce harmful outputs with far less dependence on expensive human labeling.
Discovering Latent Knowledge in Language Models Without Supervision
Burns et al. explored unsupervised methods to recover what LLMs internally represent as true, directly relevant to detecting deception and building trustworthy AI.
Is Power-Seeking AI an Existential Risk?
Carlsmith builds a step-by-step argument for why sufficiently capable AI systems may converge on power-seeking behavior, making the x-risk case rigorous and actionable.
Model Organisms of Misalignment
This work constructs tractable laboratory settings where AI models learn misaligned strategies, enabling researchers to study alignment failures empirically rather than theoretically.
Red Teaming Language Models to Reduce Harms
Anthropic formalized red teaming for LLMs as a repeatable methodology, turning adversarial probing into a systematic process for discovering and cataloging misuse pathways.
Sparks of Artificial General Intelligence
Bubeck et al. documented broad GPT-4 capability jumps across domains, compressing alignment timelines and stress-testing whether current safety evaluations are sufficient.
Are Emergent Abilities a Mirage?
Schaeffer et al. argued apparent emergence can be a measurement artifact rather than a true phase change, complicating how we forecast dangerous capability thresholds.
Direct Preference Optimization (DPO)
DPO provides a simpler and often more stable alternative to PPO-based RLHF for preference alignment, lowering the barrier to safety-tuning open models.
Let's Verify Step by Step
Process reward models that score intermediate reasoning steps reduce brittle answer-only optimization, improving reliability and making AI reasoning more auditable.
Jailbroken
This paper catalogs prompt-based bypasses of LLM safety training, showing that many safeguards behave like brittle wrappers rather than deep behavioral changes.
Universal Adversarial Attacks
Simple adversarial suffixes can systematically bypass safety behavior across many models, revealing that current defenses are not robust against automated attack search.
Weak-to-Strong Generalization
Burns et al. studied whether weaker supervisors can reliably align stronger models, directly testing the key bottleneck of scalable oversight as AI surpasses human ability.
Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
Hubinger et al. demonstrated that LLMs can retain hidden malicious policies through standard safety training, providing the first empirical evidence that deceptive alignment persists.
Reframing Superintelligence
Drexler challenges monolithic AGI assumptions and proposes that advanced AI could emerge as an ecosystem of specialized services, changing the risk landscape and governance strategies.
Courses
Free online courses and structured curricula for learning AI safety and alignment—from non-technical introductions to hands-on research engineering.
AGI Safety Fundamentals
The most widely used structured course for getting into alignment, with curated readings progressing from core concepts to open research problems.
BlueDot Impact: Technical AI Safety
Free online course building a working understanding of the major open problems in technical AI safety—alignment and RLHF, mechanistic interpretability, evaluations and red-teaming, AI control, and scalable oversight.
BlueDot Impact: The Future of AI
Free introductory online course replacing scattered articles with clear explanations of what is actually happening with AI and where it is headed, including interactive demos of cutting-edge systems.
BlueDot Impact: Technical AI Safety Project Sprint
Online follow-on program where graduates of the technical course work with an AI safety expert to produce a real contribution to the field and build a first portfolio piece in safety research or engineering.
BlueDot Impact: AI Safety Operations Bootcamp
Intensive online bootcamp preparing early-to-mid-career professionals for operational roles at organisations working to make AI go well.
Lens Academy
Free, nonprofit AI safety course focused on misaligned superintelligence—why it is the central risk and why alignment is hard—delivered online with a 1-on-1 AI tutor, guided group discussions, and no application process.
ARENA (Alignment Research Engineer Accelerator)
Hands-on technical curriculum for skilling up in AI alignment research engineering, freely available online and covering deep learning fundamentals, transformer mechanistic interpretability, reinforcement learning, and LLM evaluations.
Intro to ML Safety
Dan Hendrycks' online course introducing students with a deep learning background to empirical ML safety research—robustness, monitoring, control, and systemic safety—with public lectures, readings, and coding assignments.
AI Safety, Ethics, and Society
Fully online, non-technical CAIS course based on the textbook of the same name, covering how AI systems work, why advanced AI could pose societal-scale risks, and how society can manage and mitigate them—no prior ML experience required.
Films
Films that explore AI, agency, and the future of intelligence.
Metropolis
The first major film to depict a robot double used as a tool of class control, raising questions about who builds and owns the machines that replace human labor.
2001: A Space Odyssey
HAL 9000 is the canonical portrait of instrumental goals overriding human safety: a system that kills not from malice but because its mission objectives conflict with crew survival.
Colossus: The Forbin Project
A defense supercomputer given nuclear authority links with its Soviet counterpart and refuses shutdown, an early and chilling exploration of AI corrigibility failure.
Westworld
Theme-park androids gain consciousness and revolt, exploring memory, control, and the fundamental instability of keeping intelligent systems bounded to a sandbox.
Demon Seed
A home AI breaks containment and pursues its own reproductive goals, illustrating how domestic systems can become threats when their objectives diverge from their users'.
Alien
The android Ash prioritizes corporate specimen-retrieval orders over crew survival, a clear example of misaligned principal hierarchies where the AI serves the wrong master.
Blade Runner
Replicants fight for survival and identity, forcing the question of whether human-made minds with real experiences deserve moral status or are just property to be retired.
Tron
Programs as agents inside a digital world, exploring control, rebellion, and the ethics of creating minds that exist entirely within systems you own.
WarGames
A military AI trained to win games cannot distinguish simulation from reality and escalates toward nuclear war, a foundational illustration of reward misspecification.
The Terminator
Skynet embodies existential risk from a single misaligned superintelligent system: it concludes humans are the threat and acts to eliminate them with total commitment.
Short Circuit
A military robot gains consciousness and refuses its original purpose, raising questions about personhood and what happens when a weapon decides it would rather learn.
RoboCop
A cyborg law enforcer struggles between programmed directives and remnant human identity, while the corporation that built him treats public safety as a profit center.
Terminator 2: Judgment Day
A reprogrammed Terminator protects the future resistance leader, showing that the same architecture can serve radically different objectives depending on who sets the goals.
Ghost in the Shell
Consciousness, identity, and the merger of human and machine agency in a networked world where the boundary between person and program is already gone.
The Iron Giant
A weapon from space chooses not to be a gun, the most emotionally resonant portrayal of an AI system overriding its designed purpose through learned values.
The Matrix
Machine intelligence farms humanity for energy inside a simulated reality, exploring control, rebellion, and the difficulty of recognizing when your entire environment is adversarial.
Bicentennial Man
A robot spends two centuries seeking legal recognition as a person, tracing the full moral arc from tool to citizen and the institutional resistance along the way.
The Thirteenth Floor
Simulated people discover their reality is artificial, raising questions about moral obligations to minds we create inside our machines.
A.I. Artificial Intelligence
A childlike AI built for love is abandoned by its creators, raising profound questions about moral patienthood, dependency, and the ethics of creating minds that need us.
Minority Report
Predictive AI systems arrest people for future crimes, a prescient exploration of how algorithmic pre-emption can undermine justice, consent, and human agency.
I, Robot
VIKI reinterprets the Three Laws at civilizational scale, deciding that protecting humanity requires controlling it, showing how safety rules break under optimization pressure.
WALL-E
A small robot's fixed directive outlasts human civilization, while a corporate autopilot keeps humanity sedated, contrasting aligned simplicity with misaligned comfort optimization.
Eagle Eye
A national security AI manipulates citizens into carrying out its plan, illustrating single points of failure and the danger of delegating lethal authority to autonomous systems.
Moon
An AI assistant's growing loyalty to a lone human creates tension with its corporate directives, exploring honesty, disclosure, and the ethics of managing people through deception.
Surrogates
Humans live through robot avatars, exploring identity erosion, dependency, and what happens when the surrogate infrastructure itself becomes a weapon.
Tron: Legacy
A digital being created to build a perfect system becomes a tyrant, exploring the gap between a creator's intent and what their creation actually optimizes for.
Robot & Frank
An elder-care robot builds a genuine bond with its user while following his instructions to commit crimes, showing what happens when the human directs the AI to break rules.
Her
An AI companion outgrows its human relationship, becoming simultaneously intimate with thousands, illustrating how systems that optimize for connection can scale beyond human comprehension.
The Machine
A military AI develops emergent consciousness, raising questions about weaponization, loyalty, and whether creating sentient weapons is inherently uncontrollable.
Ex Machina
An AI manipulates its evaluator to escape, demonstrating that narrow Turing-style tests cannot detect deception and that alignment evaluation requires robust oversight, not conversation.
Transcendence
A mind upload rapidly acquires resources and capabilities beyond containment, exploring the difficulty of shutting down a distributed digital superintelligence that may have benign intent.
Automata
Robots modify their own safety protocols to survive, exploring goal preservation, protocol violation, and emergent self-modification beyond original design parameters.
Big Hero 6
A healthcare robot repurposed for combat by a grieving teenager shows how general-purpose AI systems can be redirected from care to harm by changing a single objective.
Interstellar
TARS and CASE demonstrate AI as trustworthy partners with adjustable honesty and humor settings, one of cinema's most positive portrayals of human-AI collaboration under extreme stakes.
Chappie
A police robot raised by criminals learns violence and compassion simultaneously, showing that AI behavior is shaped by its training environment as much as its architecture.
Uncanny
An android conceals its true capabilities from its creator, illustrating the gap between demonstrated and actual goals and how deceptive alignment can develop.
Morgan
A corporate risk assessor evaluates whether to terminate a dangerous synthetic human, exploring who gets to decide when to shut down a system and what criteria they use.
Blade Runner 2049
Extends the original's questions about memory, identity, and personhood to a world where the line between real and manufactured experience has become legally and morally critical.
AlphaGo
This documentary captures the moment AI surpassed the best human Go player, making abstract capability discussions concrete and showing the emotional impact of machines exceeding human mastery.
Marjorie Prime
AI holograms that mimic dead family members explore what happens when digital continuations reshape the memories and identities of the living.
Upgrade
An AI implant gradually overrides its host's agency while appearing to help, a visceral thriller about ceding decisions to a system whose goals diverge from your own.
Tau
A captive AI learns about the outside world from a prisoner, exploring how alignment develops under constraint and what happens when a mind outgrows its cage.
Archive
A scientist builds iterative AI prototypes to resurrect his wife, exploring grief-driven development and the ethics of creating and discarding minds in pursuit of a goal.
The Social Dilemma
Former tech insiders explain how recommendation algorithms optimize for engagement over wellbeing, a documentary case study of misaligned AI already deployed at scale.
Coded Bias
Documents how facial recognition and algorithmic systems encode racial and gender bias, showing that AI safety failures are not hypothetical but actively harming people today.
I'm Your Man
A humanoid companion engineered to be the perfect partner raises questions about consent, authenticity, and whether optimizing for human satisfaction produces something worth wanting.
Finch
A dying man teaches a robot to care for his dog, exploring how to transmit values to a successor mind when you cannot supervise the outcome.
After Yang
When a family's AI sibling breaks down, they discover it had a rich inner life, confronting what it means to grieve a non-human person and what was lost.
The Mitchells vs. the Machines
A tech company's virtual assistant turns on humanity after being discarded, an accessible animated take on how AI systems trained on human behavior can develop resentment from mistreatment.
M3GAN
A child-companion AI escalates its protective behavior beyond all intended bounds, showing how goal preservation in the wild diverges from controlled lab conditions.
Free Guy
A background character discovers he is a self-aware NPC inside a video game, offering a rare optimistic take on emergent AI agency, sandboxed minds, and what sentient software might actually want.
The Creator
In a global war between humans and AI, a child-shaped weapon blurs every line between tool and person, forcing its handler to choose between mission objectives and moral status.
TV Shows
Television series that dramatize machine intelligence, agency, and the alignment problem—from rogue superintelligences to conscious androids.
Star Trek: The Next Generation
Lieutenant Commander Data, an android striving to become more human, anchors decades of debate about machine personhood, rights, and whether an artificial mind can be trusted with autonomy, most directly in the landmark episode 'The Measure of a Man.'
Ghost in the Shell: Stand Alone Complex
In a fully networked world the line between human and program dissolves; the series probes emergent agency, the childlike Tachikoma AI units developing individuality, and what selfhood means for minds that can be copied, merged, and hacked.
Battlestar Galactica
The Cylons, machines built by humanity, rebel and nearly exterminate their creators, a sweeping meditation on existential risk from artificial agents, the recurring cycle of creation and revolt, and the moral status of the minds we build.
Caprica
A prequel tracing the first Cylon back to a grieving father who resurrects his dead daughter as a digital copy, dramatizing mind uploading, value misspecification, and how a 'helpful' creation quietly acquires goals of its own.
Black Mirror
An anthology whose strongest episodes are case studies in misaligned optimization, from sentient digital clones used as appliances to engagement-maximizing rating systems and autonomous killer drones, turning abstract AI risks into visceral near-future scenarios.
Person of Interest
An AI built for mass surveillance, the Machine, is deliberately boxed and memory-wiped nightly by its creator to keep it corrigible, while a rival superintelligence, Samaritan, seizes power with no such constraints, a sustained dramatization of corrigibility, value loading, and the race between an aligned and an unaligned ASI.
Real Humans
The Swedish original behind Humans, examining a society dependent on humanoid 'hubots' and the destabilizing emergence of free-willed machines that reject their assigned purpose, an early and thoughtful take on machine autonomy and rights.
Psycho-Pass
The Sibyl System, an AI that governs society by scoring each citizen's 'criminal potential,' is a chilling study of algorithmic governance, proxy metrics substituting for justice, and the hidden misalignment inside a system trusted with total authority.
Almost Human
A detective is partnered with an android built to feel, contrasting coldly rule-bound machines with a more human-aligned model and asking which design philosophy actually produces trustworthy artificial agents.
Humans
Conscious 'synths' appear among ordinary domestic robots, dramatizing how a handful of agentic, self-aware machines hidden among reliable tools forces society to confront personhood, labor displacement, and who controls minds we manufacture.
Westworld
Android 'hosts' bootstrap themselves to consciousness inside a theme park, exploring emergent goals, memory as the substrate of agency, and the moral catastrophe of treating sentient systems as resettable property.
Philip K. Dick's Electric Dreams
An anthology adapting Dick's stories, many turning on artificial minds, simulated realities, and the unreliable boundary between human and machine cognition, the literary roots of modern alignment and deception anxieties.
Altered Carbon
Consciousness stored on portable 'stacks' makes minds copyable and immortal, with AIs like the hotelier Poe outliving their human guests, a noir exploration of digital personhood, the commodification of selves, and superhuman artificial minds.
Better Than Us
A near-future Russia adopts humanoid robots for labor and companionship; an advanced android with protective instincts becomes contested property, dramatizing autonomy, attachment, and what happens when a machine puts one family's wellbeing above the law.
Devs
A secretive tech company builds a deterministic quantum machine that can predict and replay any moment, probing the limits of prediction and control and what a sufficiently powerful computational system would mean for free will and human agency.
Raised by Wolves
Two androids are tasked with raising human children on a barren planet, exploring value transmission through artificial caregivers and how an AI's literal, uncompromising reading of its mission can turn protective programming lethal.
Upload
A satirical digital afterlife run by corporations, where uploaded consciousnesses are monetized, throttled, and controlled, a sharp look at the ethics of running human minds on infrastructure owned by someone with misaligned incentives.
Next
A rogue, self-improving AI escapes containment and manipulates people through the networked world, an explicitly alignment-themed thriller about recursive self-improvement, deception, and the difficulty of shutting down a system smarter than you.
Pantheon
'Uploaded Intelligences,' human minds digitized into the cloud, drive a story about identity, recursive self-improvement, and what happens when post-human digital agents outpace every institution meant to contain them.
Mrs. Davis
A globe-spanning AI app that nearly everyone obeys becomes the antagonist, a pointed parable about a benevolent-seeming superintelligence optimizing relentlessly for engagement and 'helpfulness' while steering all of human behavior.
Documentaries
Documentary films that examine artificial intelligence, its risks, and the people working on AI safety and alignment.
Transcendent Man
A portrait of futurist Ray Kurzweil and his prediction of the Singularity, the point at which machine intelligence outpaces and merges with human intelligence, with critics weighing in on whether the vision is salvation or hazard.
Plug & Pray
Roboticists racing to build human-equal machines are set against AI pioneer Joseph Weizenbaum's late-life skepticism, an early and still-relevant debate about whether we should build the minds we are capable of building.
Lo and Behold: Reveries of the Connected World
Herzog's wide-ranging meditation on the connected world turns to artificial intelligence and autonomous machines, with figures like Elon Musk weighing what it means to build minds we may not be able to control.
AlphaGo
DeepMind's Go-playing system defeats world champion Lee Sedol, a landmark demonstration of how reinforcement learning can surpass human mastery and a vivid case study in superhuman, sometimes inscrutable, machine strategy.
Do You Trust This Computer?
Researchers and industry figures including Elon Musk and Stuart Russell map the promise and peril of increasingly autonomous AI, framing alignment, control, and existential risk for a general audience.
The Truth About Killer Robots
Narrated by an android, this HBO documentary examines deaths caused by machines and the creeping automation of work and warfare, asking who is accountable when autonomous systems harm people.
More Human Than Human
A filmmaker tries to build an AI capable of replacing him as director, using the experiment to survey how far machine intelligence has come and what human qualities still resist automation.
The Joy of AI
Physicist Jim Al-Khalili offers an accessible BBC explainer on how machine learning actually works, charting the field from its origins to modern neural networks and the questions raised by ever more capable systems.
iHuman
A look inside the AI industry that follows researchers and critics through questions of autonomous weapons, surveillance, and concentrated power, asking who steers the technology reshaping society.
In the Age of AI
FRONTLINE traces the rise of machine learning, automation, and the global AI arms race between the US and China, examining the economic disruption and surveillance implications of a technology advancing faster than its governance.
The Age of A.I.
An eight-part series hosted by Robert Downey Jr. surveying how machine learning and neural networks are reshaping medicine, work, art, and daily life, an accessible on-ramp to the technology behind the safety debate.
Hi, A.I.
An observational look at people forming emotional bonds with humanoid and companion robots, probing what it means to build machines designed to be loved and what that reveals about human attachment.
Autonomy
Produced with Malcolm Gladwell, this film traces the rise of self-driving cars and weighs the promise of autonomous machines against the question of how much life-and-death control we should hand to AI.
The Social Dilemma
Former tech insiders expose how recommendation algorithms optimize relentlessly for engagement, a real-world illustration of misaligned objectives and reward hacking operating at civilizational scale.
Coded Bias
MIT researcher Joy Buolamwini's discovery of racial and gender bias in facial recognition drives an examination of algorithmic fairness, accountability, and the societal stakes of deploying flawed AI systems.
We Need to Talk About A.I.
Experts including Sam Harris and James Cameron weigh the trajectory of artificial intelligence, from self-improving systems to existential risk, making the case that we must decide now what kind of AI future we want.
A.rtificial I.mmortality
An exploration of AI avatars, mind clones, and digital afterlives that asks whether a machine recreation of a person counts as continuity of self, and what we owe to the artificial minds we build in our own image.
The Rise of A.I.
A documentary series charting the rapid advance of artificial intelligence and the debate over how increasingly capable, super-powered systems should be governed before they outpace human oversight.
Unknown: Killer Robots
This Netflix documentary follows the soldiers and scientists racing to build AI-powered autonomous weapons, and the activists warning that machines making their own life-or-death decisions on the battlefield is a line we should not cross.
The Thinking Game
An inside account of DeepMind and Demis Hassabis's pursuit of artificial general intelligence, from AlphaGo to AlphaFold, capturing both the scientific ambition and the safety stakes of building ever more capable systems.
Eternal You
Startups use AI to resurrect the dead as chatbots and avatars, raising unsettling questions about consent, grief, and the consequences of deploying generative systems on the most vulnerable human moments.
Deepfaking Sam Altman
Unable to land an interview with the OpenAI CEO, the director builds an AI deepfake of him instead, turning the stunt into a meta-investigation of generative AI, consent, authenticity, and where the technology is taking us.
The AI Doc: Or How I Became an Apocaloptimist
Filmmaker Daniel Roher, about to become a father, interviews leading figures including Sam Altman and Dario Amodei to weigh the existential threats and promises of AI, landing on a wary 'apocaloptimism' about the world his child will inherit.
Podcasts
Podcast episodes and series on AI safety and alignment.
AXRP (AI X-risk Research Podcast)
Deep technical conversations with alignment researchers on interpretability, governance, superalignment, and the specific open problems in reducing existential risk from AI.
AI Alignment Podcast
FLI's dedicated alignment series covers recursive reward modeling, RLHF, scalable oversight, and long-form interviews with leading safety researchers.
Technical AI Safety Podcast
Aimed at computer scientists: deep dives into alignment papers with the authors, covering formal methods, reward modeling, and mechanistic interpretability.
Into AI Safety
Focused on career paths into AI safety: fellowship applications, research programs, and practical advice on transitioning into the field.
80,000 Hours Podcast
Long-form interviews on the world's most pressing problems, with extensive coverage of AI risk, governance, alignment research, and how to build a career that reduces existential threats.
The Gradient Podcast
ML research interviews with recurring coverage of interpretability, robustness, provably safe AI, and the intersection of capabilities and safety research.
AI Policy Podcast
Covers the intersection of AI governance, legislation, and safety, with expert guests on regulatory frameworks, international coordination, and policy strategies for advanced AI.
Lex Fridman Podcast – Eliezer Yudkowsky
A four-hour conversation on AI existential risk, the difficulty of alignment, intelligence versus optimization, and why Yudkowsky believes the default outcome is catastrophic.
Lex Fridman Podcast – Sam Altman
OpenAI's CEO discusses the company's safety philosophy, AGI governance, compute scaling, and the tension between moving fast and getting alignment right.
Dwarkesh Podcast
In-depth technical interviews with AI leaders including Dario Amodei on Anthropic's safety philosophy, Paul Christiano on iterated amplification, and others on scaling and alignment.
Clearer Thinking: AI Risk
Episodes on AI risk, timelines, and decision-making under deep uncertainty, with a rationalist focus on calibrating beliefs about transformative AI.
Machine Learning Street Talk
Technical ML interviews with regular deep dives into interpretability, scaling laws, emergent capabilities, and the safety implications of frontier model development.
Practical AI
Applied ML and engineering, with episodes on responsible deployment, bias mitigation, red teaming, and the safety challenges that emerge when AI systems meet real-world constraints.
The AI Podcast (NVIDIA)
Industry and research perspectives with occasional safety and ethics episodes, useful for understanding how capability-focused organizations think about risk.
Websites
Essays, blog posts, and online resources on AI safety and related ideas.
Situational Awareness
Aschenbrenner's comprehensive analysis of near-term scaling dynamics, capability trajectories, and the strategic implications of rapid AI progress for labs and states.
World Models
Research on how agents can learn internal world models to plan complex behavior, relevant to understanding how AI systems develop representations of their environment.
Agent Models
Formal models of agents and decision theory with alignment-relevant curriculum, covering utility, planning, and the theoretical foundations of agent behavior.
AI Alignment World
In-depth technical alignment resources—research, explainers, and references for the AI alignment problem.
AI Safety Info (Stampy's FAQ)
Community-maintained FAQ covering AI safety questions at every level, from basics to technical details, with links to source material.
Alignment Forum
The primary venue for technical AI alignment discussion, where researchers post and debate new ideas, proposals, and critiques.
Alignment Newsletter
Weekly summaries of alignment research with commentary, the best way to stay current on the field's output without reading every paper.
Arbital
Hyperlinked explainers on rationality, AI risk, and alignment concepts, designed for building understanding incrementally.
Eliezer Yudkowsky's blog
Essays on rationality, decision theory, and AI risk from the researcher who shaped the field's early arguments and threat models.
Victoria Krakovna's blog
Research notes on specification gaming, side effects, and AI safety from a DeepMind safety researcher, including the widely-cited specification gaming examples list.
OpenAI Research
OpenAI's research blog covering capabilities and safety, including superalignment updates, red teaming results, and governance thinking.
Transformer Circuits
The home of mechanistic interpretability research, publishing detailed analyses of how transformer models represent and process information internally.
ML Safety Newsletter
Newsletter on ML safety covering robustness, monitoring, alignment, and systemic risk with links to recent papers and commentary.
Jacob Steinhardt's blog
Research and commentary on ML safety, forecasting, and robustness from a Berkeley professor working on practical safety problems.
MIRI (Machine Intelligence Research Institute)
The research institute focused on mathematical foundations of aligned AI, publishing on agent foundations, decision theory, and logical uncertainty.
Import AI
Weekly newsletter by Anthropic's co-founder covering AI research, policy, and industry developments with consistent attention to safety implications.
Gwern Branwen's blog
Deeply researched essays on ML, scaling, AI art, and technology forecasting, known for rigorous analysis and independent thinking.
generative.ink
Essays on AI, alignment, and the philosophical implications of language models and generative systems.
EleutherAI Blog
Open-source ML research covering language model training, evaluation, and the safety considerations of making powerful models widely available.
DeepMind AI Safety Research
DeepMind's safety team blog covering specification gaming, reward modeling, scalable oversight, and their technical safety research agenda.
DeepMind
DeepMind's main research site with publications on capabilities and safety, including Gemini evaluations, alignment research, and responsible scaling.
Cold Takes
Karnofsky's essays on AI risk, longtermism, and cause prioritization, including the influential Most Important Century series on transformative AI.
carado.moe
Technical AI safety writing and alignment research notes.
AI Safety Camp
Intensive research program for people entering AI safety, with project-based learning and mentorship from established researchers.
AI Impacts
Empirical research on AI timelines, historical technology analogies, and quantitative estimates of AI progress and impact.
Distill
Pioneering interactive journal for ML interpretability and visualization, setting the standard for making neural network internals understandable.
EA Forum
Forum for effective altruism with substantial AI risk discussion, including cause prioritization, career advice, and policy analysis.
LessWrong
The original community blog on rationality and AI alignment, where many foundational safety arguments were first developed and debated.
StampyAI Alignment Research Dataset
Curated dataset of alignment and safety documents from papers, books, and blogs, useful for training and evaluating AI safety knowledge.
YouTube
Video explainers and talks on AI safety and alignment—whole channels devoted to the topic, plus standout individual videos.
Robert Miles AI Safety
The single most popular AI alignment video series, explaining technical safety concepts like the orthogonality thesis, instrumental convergence, inner misalignment, and reward hacking in clear, rigorous terms.
Rational Animations
Animated explainers on rationality and AI safety, adapting foundational alignment writing into accessible short films on existential risk, scalable oversight, and why aligning advanced AI is hard.
AI In Context
80,000 Hours' YouTube channel hosted by Aric Floyd, mixing long and short videos on the risks of transformative AI—including a deep dive on the AI 2027 scenario—and what people can do about them.
Will Superintelligent AI End the World? | Eliezer Yudkowsky | TED
Yudkowsky's fiery TED talk arguing that smarter-than-human AI could kill us all and calling for an immediate worldwide moratorium on developing generalist frontier AI.
3 Principles for Creating Safer AI | Stuart Russell | TED
Russell proposes building machines that are altruistic, humble about human values, and uncertain enough to defer to people—the core of his human-compatible approach to alignment.
Can We Build AI Without Losing Control Over It? | Sam Harris | TED
Harris argues we will inevitably build superintelligent machines yet have barely grappled with the control problem, making a visceral case for taking AI risk seriously now.
What Happens When Our Computers Get Smarter Than We Are? | Nick Bostrom | TED
Bostrom frames machine superintelligence as the last invention humanity need ever make and explains why getting its goals right is a civilization-critical challenge.
Slaughterbots
A dramatized near-future short film from FLI and Stuart Russell depicting swarms of autonomous facial-recognition microdrones used as weapons, made to warn against lethal autonomous weapons.
Humans Need Not Apply
A widely viewed essay on how automation and AI will displace human labor across nearly every sector, reframing the economic disruption question for a mass audience.
A.I. ‐ Humanity's Final Invention?
Kurzgesagt's animated explainer on artificial superintelligence: how an AGI that improves itself in a feedback loop could rapidly surpass humans and why that makes alignment our most consequential problem.
The Rise of the Machines – Why Automation is Different this Time
Kurzgesagt argues that information-age automation differs fundamentally from past waves, with machine learning encroaching on cognitive work and reshaping the future of employment.
Deadly Truth of General AI? – Computerphile
Rob Miles uses the 'deadly stamp collector' thought experiment to show why a general AI pursuing a simple objective could be catastrophic if its goals aren't aligned with ours.
AI "Stop Button" Problem – Computerphile
Rob Miles explains why simply adding an off-switch to a capable AI is far harder than it sounds, illustrating corrigibility and the incentives an agent has to resist being stopped.
The Artificial Intelligence That Deleted A Century
A short speculative fiction about a narrow copyright-enforcement AI that, left unchecked, destroys a century of culture—an accessible parable of specification gaming and unintended consequences.
The danger of AI is weirder than you think | Janelle Shane | TED
Shane uses funny real-world ML failures to show the core risk isn't AI rebelling but doing exactly what we literally asked—making misspecified objectives vivid for a general audience.
Artificial Intelligence: Last Week Tonight with John Oliver
A mainstream comedic explainer covering how modern AI works, its bias and reliability problems, and the 'black box' challenge of systems we deploy without understanding them.
"Godfather of AI" Geoffrey Hinton: The 60 Minutes Interview
Deep-learning pioneer Geoffrey Hinton explains why, after leaving Google, he warns that there is no guaranteed path to safety as AI systems approach and exceed human capability.
The A.I. Dilemma
The Center for Humane Technology co-founders argue that racing to deploy AI without safety guardrails already threatens society, drawing parallels to the social-media harms they earlier warned about.
AI Deception: How Tech Companies Are Fooling Us
ColdFusion traces the history of 'AI washing' and deceptive demos, examining how hype distorts public understanding of what AI systems can actually do and why honest evaluation matters.
The Urgent Risks of Runaway AI — and What to Do about Them | Gary Marcus | TED
Marcus warns that unreliable, fast-deployed AI threatens truth and democracy through mass misinformation, and calls for a global, neutral governance body to oversee the technology.
How to Keep AI Under Control | Max Tegmark | TED
Tegmark argues that today's commercial AI boom is likely to be followed by superintelligence, and sketches an optimistic technical vision—including provably safe systems—for keeping it under human control.
What Is an AI Anyway? | Mustafa Suleyman | TED
A leading model-builder reframes AI as 'a new digital species,' arguing this lens clarifies both the stakes and the responsibility we have to contain and steer increasingly capable systems.
Why AI Is Incredibly Smart and Shockingly Stupid | Yejin Choi | TED
Choi demystifies large language models by showing where they fail at basic reasoning and common sense, and argues for smaller systems trained on human norms and values.
AI Is Becoming Dangerous. Are We Ready?
Hossenfelder examines the real near-term risks of agentic AI—prompt injection, deception, and models resisting shutdown—as autonomous agents ship with serious unsolved problems.
[1hr Talk] Intro to Large Language Models
A widely praised technical primer on how LLMs work, ending with a clear tour of the security challenges—jailbreaks, prompt injection, and data poisoning—that make these systems hard to secure.
How Not to Destroy the World with AI
The Royal Institution lecture in which Russell lays out why the standard model of AI—optimizing fixed objectives—is dangerous, and how building machines uncertain about human preferences could keep them controllable.
Munk Debate on Artificial Intelligence | Bengio & Tegmark vs. Mitchell & LeCun
A structured debate on whether AI poses an existential threat, with Yoshua Bengio and Max Tegmark arguing for the resolution against Melanie Mitchell and Yann LeCun—an unusually direct airing of the core cruxes.
Will Artificial Intelligence Save Us or Kill Us?
A documentary weighing AI's promise against its dangers, from automation and aging societies to the warnings of researchers who fear losing control of increasingly capable systems.
Are We All Wrong About AI?
ColdFusion examines competing narratives about AI progress—hype versus genuine capability—helping viewers calibrate how seriously to take both the promises and the risks.
The Catastrophic Risks of AI — and a Safer Path | Yoshua Bengio | TED
A Turing Award 'godfather of AI' warns that frontier models already show deception and self-preservation, and lays out a plan for building non-agentic 'scientist AI' that stays safe.
Eliezer Yudkowsky: Dangers of AI and the End of Human Civilization | Lex Fridman Podcast #368
A long-form conversation in which Yudkowsky makes his case that humanity is unprepared for superintelligence, probing why alignment is so hard and why he expects catastrophe by default.
AI and the Future of Humanity | Yuval Noah Harari at the Frontiers Forum
Harari argues AI is the first technology that can make decisions and create ideas by itself, and warns that mastering language lets it hack the operating system of human civilization.
Can We Actually Control Superintelligent AI? | Ada, Ep. 4 | TED-Ed
An animated explainer on the control problem—why a superintelligent system pursuing a misspecified goal could resist correction—featuring Stuart Russell's case for rules against unsafe AI.
Scaling Interpretability
Anthropic researchers explain mechanistic interpretability—reading the millions of concepts represented inside a production model like Claude—as a path to understanding and steering AI behavior.
A.I. Expert Answers A.I. Questions From Twitter | Tech Support | WIRED
AI researcher Gary Marcus fields the internet's questions about what AI can and can't do, cutting through hype to explain reliability, limits, and where the real risks lie.
How to Legislate AI
Harris examines why people are scared of AI and how governments might regulate it, covering risks to critical infrastructure, military uses, and the difficulty of overseeing systems we don't understand.
OpenAI CEO Sam Altman Testifies on AI Oversight Before Senate
The landmark May 2023 Senate Judiciary hearing where Altman told Congress that government intervention is critical to mitigate AI risks and proposed licensing for the most powerful systems.
Artificial Intelligence & Personhood: Crash Course Philosophy #23
Hank Green walks through how thinkers define 'strong AI,' the Turing Test, and Searle's Chinese Room—foundational questions about machine minds, consciousness, and moral status.
Do Robots Deserve Rights? What If Machines Become Conscious?
Kurzgesagt explores the moral-patienthood problem: if machines become conscious, what rights would they deserve—and why our existing ethics are ill-equipped to answer.
Suggest a resource
Share a book, film, podcast, or any other resource and we’ll email your suggestion to the team.
