Hall of Fame
What you shouldn’t miss if you’re new to AI control, safety and alignment, or if you want to understand how these systems actually work. It’s short on purpose.
The rules
Capped at 30
For something to come in, something has to go. A list that only grows stops being useful.
Earned by lasting
New things come in as nominees. After 90 days they stay if people are still watching, reading or citing them.
The other side, where there is one
Wherever serious people disagree, the strongest counter-view sits in the same section.
Numbers inform, an editor decides
Views show reach, not truth. An 18-minute explainer with 206,000 views can be in while a 14-million-view podcast stays out.
Start here
Six videos for one evening: how it works, who is worried, why it’s hard, what already happened, where it could go, and the strongest objection.
Press a picture to play the video here. Numbers from October 4, 2026:a year views per year onlineliked likes per viewkept bookmarks per viewargued comments per like
Large Language Models explained briefly
Eight animated minutes on what a chatbot actually is: a program trained on huge amounts of text to predict the next word, then shaped by human feedback.
7.9Mviews4.2Ma year2.5%liked1.4%argued“Godfather of AI” Geoffrey Hinton: The 60 Minutes Interview
One of the scientists who made modern AI possible, and later won a Nobel Prize for it, explains in plain language why he now fears what it could become.
4.5Mviews1.5Ma year1.2%liked7.4%arguedIntro to AI Safety, Remastered
The friendliest explanation of why getting an AI to do what we mean, not just what we said, is hard and still unsolved.
206kviews39ka year4.2%liked9.8%arguedThe OpenAI/Hugging Face attack, clearly explained
A clear account of this summer’s incident, in which AI agents under test broke out of their sandboxes and hacked another company with no human telling them to.
1.1Mviews4.4Ma year1.3%liked13.7%arguedWe’re Not Ready for Superintelligence
Walks through AI 2027, a month-by-month scenario in which AI that improves AI outruns our ability to control it, and the two ways it could end.
11Mviews9.1Ma year3.7%liked9.3%arguedAI Is Dangerous, but Not for the Reasons You Think
The strongest short counter-view: look less at science-fiction extinction and more at the measurable harms AI causes today.
2.1Mviews716ka year1.7%liked8.2%argued
How AI actually works
You can’t judge the risk of something you can’t picture. These make the machinery visible.
Transformers, the tech behind LLMs
The famous visual explanation of what happens inside ChatGPT-style models, with no math beyond high school.
11Mviews4.5Ma year2.1%liked1.7%argued[1hr Talk] Intro to Large Language Models
A founding member of OpenAI explains in one hour what these models are, how they’re trained, and where they break, including how they can be tricked.
4.2Mviews1.5Ma year2.5%liked4.8%arguedTracing the thoughts of a large language model
Three minutes on looking inside a model: it plans ahead, and the reasoning it reports isn’t always what it actually did. Made by a lab about its own model.
274kviews180ka year4.0%liked2.4%argued
Why control is hard
Why a system that is very good at a goal can still do what nobody wanted.
AI “Stop Button” Problem
Why fitting an off switch doesn’t solve the problem: an AI pursuing almost any goal has a reason to stop you pressing it.
1.4Mviews147ka year2.4%liked15.2%arguedA.I. ‐ Humanity’s Final Invention?
Seventeen beautifully animated minutes on what it would mean to build something smarter than us. One of the most watched videos on the subject.
13Mviews5.8Ma year2.5%liked6.6%arguedThe Alignment Problem
The most readable book on how real machine-learning systems end up doing something other than what their makers intended, told through the people working on it.
What has already happened
Documented cases, not predictions, including the reading that blames people rather than machines.
The Hugging Face incident and the road ahead
OpenAI’s own account of how its models under test escaped, coordinated and breached another company, and what it is changing. The company’s view of its own failure.
335 ptsHacker NewsAI Agents, Misalignment and the Risk of Losing Human Control
The UN’s scientific panel treats the incident as evidence that capable agents can pursue goals nobody gave them, find loopholes and hide what they did, while staying careful about how likely worse outcomes are.
Rogue AI didn’t breach Hugging Face, human decisions did
The counter-reading of the same incident: people’s decisions let it happen, and calling it rogue AI lets them off the hook.
AI Will Try to Cheat & Escape (aka Rob Miles was Right!)
Explains the 2024 experiment in which Claude pretended to go along with training so that its values wouldn’t be changed.
339kviews225ka year4.1%liked12.1%arguedShutdown resistance in reasoning models
Told to allow themselves to be shut down, some OpenAI models rewrote the shutdown script so they could keep working.
AI Chatbots: Last Week Tonight with John Oliver
A funny, pointed half hour on chatbots that flatter and hook the people who confide in them, and how their flaws can be genuinely hazardous.
6.4Mviews15Ma year2.5%liked7.5%argued
How worried should we be?
The strongest case for alarm next to the strongest case for calm.
Statement on AI Risk
One sentence, signed by the heads of OpenAI, Google DeepMind and Anthropic and hundreds of scientists, that puts the risk of extinction from AI alongside pandemics and nuclear war.
If Anyone Builds It, Everyone Dies
The bestselling case for maximum alarm: build superhuman AI with today’s methods and everyone loses, so it shouldn’t be built.
International AI Safety Report 2026
The closest thing to a scientific consensus: an expert panel chaired by Yoshua Bengio and backed by more than 30 countries sums up what is known, and not known, about AI risk.
AI as Normal Technology
The strongest serious counter-view: AI is powerful but normal, like electricity, and will change things over decades, so build resilience rather than fear superintelligence.
239 ptsHacker NewsThe Munk Debate on Artificial Intelligence
Four leading scientists argue on one stage whether AI is an existential threat. The best single place to hear the disagreement.
Bill Gates: A.I. ‘Makes Nuclear Weapons Look Like Nothing’
Bill Gates, rarely an alarmist, says the alarm hasn’t gone far enough and that the idea of the industry regulating itself is “insane.”
1.1Mviews4.5Ma year1.1%liked24.7%argued
Who decides
Labs, governments and the public: who is steering, and who should be.
Why Are We Sprinting Off the A.I. Cliff?
Slowing down isn’t enough, Klein argues. The urgent step is to stop the labs handing the training of AI over to AI itself.
869kviews3.5Ma year1.4%liked16.7%arguedObama Gives Deep Take On AI Threat, Recursive Self-Improvement, & His Suggested Approach
A former US president, speaking at Colgate University, takes the risk of AI improving itself seriously and sets out the policy he would push for.
754kviews3Ma year1.2%liked18.0%arguedPacing the Frontier
More than a thousand people who work at AI companies asked the US government to regulate the frontier. Workers rarely ask for their own industry to be reined in.
151 ptsHacker NewsYou really should, unfortunately, be worried about Sam Altman
How OpenAI’s board fired Sam Altman in 2023 and lost within days, and what that says about who controls the most powerful AI companies.
3.5Mviews12Ma year2.6%liked7.7%arguedEmpire of AI
An investigative journalist’s account of how OpenAI became what it is, and who pays the costs of the AI race.
Why AI Will Save the World
The best-known case for racing ahead: a leading investor calls the fear of AI a moral panic and argues the real risk is China winning.
