image generation

Alex O'Connor shows that ChatGPT cannot draw a wine glass filled to the brim, and ties the failure to David Hume's missing shade of blue. It asks what the failure says about how image models learn.

A guest video by Welch Labs on 3Blue1Brown explaining how text-to-image and video models work. It covers CLIP, shared embedding spaces, diffusion models (DDPM, DDIM), conditioning, guidance and negative prompts.

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.

Veritasium visits the US Navy's Crane facility to test top night vision goggles. It explains how image intensifier tubes and thermal or digital systems work and compares the generations of the technology.

Benn Jordan's video about storing a PNG image in a living bird, a starling that learns to reproduce encoded sounds. It mixes ornithology, audio engineering, and data encoding.

MinutePhysics explains that computers store brightness nonlinearly, so blurring and blending colors in raw RGB gives wrong, dark results. It shows why mixed colors, such as in gradients, look off.

Martin Donald explains the Wave Function Collapse algorithm using Sudoku-like constraint solving and shows how he implemented it with Blender and Godot. It links demos and reference articles.

A short Techquickie explainer on why laptop webcams are much worse than phone cameras. It covers sensor size, image processing, and the interface differences between laptops and phones, with input from a Logitech engineer.

A short animated explainer that stages a $100 footrace in which contestants start at different distances from the line, depending on the advantages or disadvantages they inherited. It uses the race as a metaphor for how unequal starting points shape outcomes, citing US census income and poverty figures.

Vsauce looks at 'juvenoia', the exaggerated fear that new media and youth culture are corrupting the young. It gives historical examples of older generations complaining about the next one.

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

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.

Yannic Kilcher walks through the Attention Is All You Need paper, which introduced the Transformer, a sequence model based solely on attention without recurrence or convolutions. He reviews the architecture and the paper's translation results.

Rob Miles explains on Computerphile how language models generate plausible text and how transformers improved on recurrent networks, with a look at what comes next.

A short cartoon explaining how language models like ChatGPT work, from autocomplete and conditional probabilities to neural networks, gradient descent and backpropagation. It shows how networks act as universal function approximators.

Kurzgesagt examines how AI-generated content is flooding the internet, from video and music to news and books. It looks at how this makes verifying what is real harder.

A Welch Labs explainer on neural scaling laws and the compute-efficient frontier, asking why model loss seems to hit a line it cannot cross. It draws an analogy to an ideal gas law for AI.

Rob Miles explains "glitch" tokens, odd strings in large language model vocabularies that make the model produce gibberish. He covers how they likely arise from tokenizer training data, such as Reddit usernames, that never appeared in the model's own training.

Musician Benn Jordan explains and demonstrates adversarial noise that poison-pills music files against AI training and scraping.

Chubbyemu presents a case study of a man who followed AI health advice, replaced salt with sodium bromide and developed bromism with severe neurological effects. It is told as a dramatized medical case by a doctor.

Musician Benn Jordan uses AI to detect AI-generated music and digs into music industry data, including streaming payouts and sample library copyright problems. He covers ways to tell whether a track is human-made.

Computerphile video in which Mike Pound explains a University of Maryland paper on watermarking LLM output. The method biases token choices so generated text can later be detected statistically.

Rob Miles discusses GPT-3 on Computerphile, covering how scaling parameters kept improving results over GPT-2 and how the model nearly learned arithmetic without anyone knowing exactly how.

Mike Pound explains the discrete cosine transform and how it underpins JPEG compression. It is part two of Computerphile's JPEG series.

Veritasium explains what a black hole should look like, using J.P. Luminet's 1979 work and a cardboard 3D model to show how light bends around it. It was made just as the first black hole image was released.

A short video built around an animation of the roughly 10,000 galaxies in the Hubble Ultra Deep Field, placed at their measured redshift distances. It gives a brief history of deep field images and ends with a 3D fly-through.

Explains the proposed NASA JPL plan to use the Sun as a gravitational lens, sending solar sail craft to its focal region to image an exoplanet's surface. It covers the distances involved, how imaging with the Sun would work and the mission design.

Practical Engineering's explanation of the February 2021 Texas winter storm and the resulting power grid outage. It lays out the facts about ERCOT, the grid's isolation, and how generation across fuel types failed in the cold.

PBS Space Time explains the proposal to use the Sun's gravity as a giant lens, sending a train of small light-sail spacecraft to the focal point roughly 550 AU out. It covers how such a telescope could image exoplanets in surprising detail and the engineering challenges involved.

Practical Engineering explains why connecting solar and other inverter-based generation to the grid is hard. It covers grid stability, load balancing and the behind-the-scenes engineering challenges of renewables.

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 explains stellar engines such as the Shkadov thruster and Caplan thruster, which could move the Sun and solar system across the galaxy over millions of years to avoid hazards and reach new regions.

3Blue1Brown explains space-filling curves and the Hilbert curve, and how infinite mathematical constructions connect to finite, practical approximations. It touches on mapping between dimensions and sensory ideas.

Wes Bos breaks down the techniques behind the McMaster-Carr website's speed: server-rendered HTML, prefetching, CDN and service worker caching, preloading, critical CSS and fixed-size images. He also notes its old-school jQuery and YUI JavaScript.

A re-upload of a video breaking down Eminem's rhyme schemes, showing multi-syllable and internal rhymes in songs such as Lose Yourself. It aims to show why he is considered a standout lyricist.

The Slow Mo Guys film a TV and screens with a high-speed camera to show how displays build an image. The footage includes CRT-style scanning and how pixels light up over time.

Technology Connections explains the difference between power (watts) and energy (watt-hours) and why confusing them leads to errors. It runs about 52 minutes and uses everyday examples.

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.

A Two Minute Papers episode on NVIDIA's recurrent denoising autoencoder, which cleans up noisy Monte Carlo path-traced image sequences. It shows how a neural network can produce usable renders from very few samples.

Retro Game Mechanics Explained shows how the Atari 2600 draws images with almost no video memory, by having the CPU race the CRT beam line by line. It walks through the techniques in detail.