ear training

Yosh trains a reinforcement learning agent to drive Trackmania until he can no longer beat it. The video explains how the training works and the design choices involved.

Matt Parker of Stand-up Maths demonstrates MENACE, Donald Michie's 1961 machine made of matchboxes that learns to play noughts and crosses through reinforcement. He builds and trains a physical version.

3Blue1Brown's second neural network chapter explains cost functions and gradient descent, showing how a network learns by minimizing error during training.

David and Ben Crystal introduce the Original Pronunciation production of Shakespeare and show what it reveals about the history of English, including puns and rhymes lost in modern speech.

3Blue1Brown's first chapter on deep learning explains what neurons and layers are and the math underlying a basic neural network, using handwritten digit recognition with visual animations.

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.

Two Minute Papers covers a paper on speeding up fluid and smoke simulation with convolutional networks. It contrasts the approach with earlier regression forest methods.

A 10-minute 3Blue1Brown chapter explaining the determinant as the factor by which a linear transformation scales area or volume. It builds the idea geometrically with animations before any formulas.

Kurzgesagt compares how many people nuclear accidents have killed against other energy sources. It walks through the worst nuclear accidents and puts their death tolls in context per unit of energy.

Kyle Hill explains what nuclear waste actually is and how it is stored, arguing that the problem was effectively solved decades ago. It covers solid shielded waste and deep geological storage.

Real Engineering looks at why nuclear power plants are expensive and how their economics compare with fossil and renewable alternatives. It is based on a video from the Illinois Energy Professors.

Scott Manley explains the Cold War experiments in which steel and aluminium spheres, known as Lew Allen's Balls, were placed near nuclear detonations. The spheres survived the fireball, and he walks through what the declassified results show.

Scott Manley explains the nuclear salt water rocket, Robert Zubrin's concept of a rocket fueled by a rapid nuclear reaction in uranium salt dissolved in water. He covers its extraordinary performance and the dangers of managing it.

Cleo Abram's explainer on whether nuclear waste is really the intractable problem it is portrayed as, looking at how spent fuel can be recycled and used. It features a collaboration with Johnny Harris.

3Blue1Brown's visual walkthrough of the attention mechanism inside transformers, covering self-attention, multiple heads and cross-attention. It is chapter 6 of his deep learning series.

3Blue1Brown's third deep learning chapter gives a visual, intuitive account of backpropagation and what happens to a neural network as it learns. A follow-up video covers the underlying calculus.

Randall Munroe of xkcd answers what would happen if you swam in a spent nuclear fuel pool, and how much radiation you would receive at the surface versus at depth.

A short Boeing-supplied clip where the 737 MAX program's chief project engineer explains how the lengthened landing gear of the MAX 10 folds up. It shows the gear model in motion.

3Blue1Brown's visual explanation of how transformers and GPT-style language models work, tracing data through tokens, embeddings and attention.

Two Minute Papers covers OpenAI's multi-agent hide-and-seek experiment, in which agents learned emergent tool use and found ways to break the game's rules. It summarizes the 'Emergent Tool Use from Multi-Agent Interaction' paper.

Sabine Hossenfelder explains how close nuclear fusion really is to break-even, and why headlines about ITER and lab milestones are misleading. She distinguishes plasma gain (Q plasma) from whole-system gain (Q total) and shows the latter is still far below one.

3Blue1Brown's chapter on eigenvectors and eigenvalues in the Essence of Linear Algebra series. It builds a visual intuition for what they mean, why an eigenbasis is useful, and ends with a puzzle.

Veritasium tells how Kodak film was fogged by radioactive fallout from the first atomic tests, letting the company detect nuclear testing before anyone announced it. The video also covers the deal to keep it quiet and how fallout contaminated packaging.

Practical Engineering explains what a 100-year flood actually means, combining statistics, hydrology and how society treats risk. Grady walks through local drainage infrastructure and the caveats in estimating flood probability.

Medlife Crisis examines why some elite endurance athletes die in their sleep, covering extremely slow heart rates and the effects of performance-enhancing drugs that thicken blood. It uses cyclist Marco Pantani as an example.

3Blue1Brown's chapter 4 on deep learning gives a more formal, symbol-based account of backpropagation calculus, connecting the intuition from the previous chapter to the math found in textbooks and code.

3kliksphilip summarizes Valve's GDC talk on VACNet, a deep-learning system that flags CS:GO cheaters and feeds the Overwatch review queue. It covers Trust Factor, VAC, and how the data is collected and analyzed.

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.

CGP Grey argues for picking a theme for the year or season instead of setting goals or new year's resolutions. He explains how a theme gives direction without a pass/fail target.

Veritasium debunks the idea of learning styles, such as being a visual learner, with an interview with Daniel Willingham and a memory experiment. It argues that a multimodal approach helps everyone.

A Smarter Every Day episode from aboard a nuclear submarine on how the crew makes breathable air. It covers oxygen generation from seawater, oxygen candles and CO2 scrubbing with amine.

Kurzgesagt's hour-long animated film that compresses Earth's 4.5 billion year history into one hour, roughly a million years per second. It follows the evolution of life through to humanity's tiny final stretch.

Vox Earworm explains why John Coltrane's Giant Steps is so hard to improvise over: it is fast and moves through three keys. Musicians Braxton Cook and Adam Neely walk the host through the underlying music theory.

Real Engineering's Brian McManus explains Helion's approach to nuclear fusion: colliding two plasma rings, compressing them magnetically, and capturing energy directly from the changing magnetic flux rather than boiling water. It covers why this pulsed design sidesteps problems of tokamaks and laser fusion.

Rick Beato argues that adults cannot develop true perfect pitch, which he ties to critical-period brain plasticity and language acquisition. He says it can only be developed in young children, and discusses the research behind this.

Primer simulates green beard altruism, where individuals recognize and help others carrying a shared marker, and shows how imposters affect its evolution. It is an animated evolution simulation based on published papers.

Veritasium visits Chernobyl, Fukushima and other places to find who on Earth is exposed to the most ionizing radiation. The answer turns out to be surprising and puts radiation risk in context.

A short explainer from Nat and Friends on where Google Earth's 3D imagery comes from, how it is stitched together, and how often it is updated. It describes aircraft carrying camera rigs and photogrammetry that turns the photos into 3D models.

Big Clive takes apart an IKEA Koppla USB power supply to assess its safety, looking at isolation and transformer construction.

Hugh Jeffreys tears down the iPhone 12 and assesses its repairability, focusing on how parts are serialized and tied to the device so swapping components between phones does not work.