Page 1 of 24
What you’ll learnHow backpropagation really works, by building a tiny autograd engine in Python line by line and then training a small neural net with it. Only basic Python and school calculus assumed.
52:52Backpropagation through a single neuron, by hand
What you’ll learnWhat a neural network actually is (neurons, layers, weights and biases), pictured through a network that learns to read handwritten digits.
8:38How layers could pick out edges and shapes
What you’ll learnWhy the old rainbow “jet” colormap misleads the eye, how perceptual colour spaces work, and how viridis was designed to replace it.
What you’ll learnThe strongest case that self-driving cars could deepen sprawl, congestion and car dependence, and what cities could choose instead.
28:04How autonomous vehicles could reshape city streets
What you’ll learnTwo hundred years of global health and wealth in four minutes, with life expectancy plotted against income for 200 countries since 1810.
What you’ll learnThe argument that automation could make much human work obsolete the way engines did for horses. Made in 2014, so it also shows how such predictions age.
What you’ll learnHow a GPT-style model turns text into tokens and embeddings and predicts the next word, drawn step by step.
12:27Word embeddings, visualised
What you’ll learnThe inside story of DeepMind’s AlphaGo facing Go champion Lee Sedol in 2016, and how the players and the team reacted when the machine won.
What you’ll learnWhy birth rates fall as child survival and incomes rise, and why world population is expected to level off, told with Hans Rosling’s animated data.
What you’ll learnA very funny but serious keynote on why trusting opaque machine learning and fast-moving tech with important decisions is harder than the industry admits.
What you’ll learnThe opening lecture of Andrew Ng’s Stanford machine learning course, which sets out supervised, unsupervised and reinforcement learning and what the course will cover.
What you’ll learnHow a program evolves neural networks with the NEAT genetic algorithm until one can finish a Super Mario World level, in six minutes.
In briefAndrej Karpathy's three-hour-plus general-audience deep dive into how ChatGPT-style LLMs are built, covering the full training stack from pretraining to post-training. It also gives mental models for their quirks, like hallucination, and how to use them well.
In briefTom7 explains and demos software that learns to play NES games by finding score patterns in memory and then searching for inputs that raise it. It is the video companion to his SIGBOVIK paper.
In briefThe first lecture of David Silver's reinforcement learning course, introducing the field and its basic concepts. The course follows the Sutton and Barto textbook.
In briefVeritasium explains why a large share of published research findings may be false, covering p-values, p-hacking, publication bias and low replication rates. It draws on John Ioannidis's well-known paper and describes reforms such as preregistration.
In briefAndrej Karpathy builds a GPT from scratch in code, following Attention Is All You Need and GPT-2/3, in about two hours using PyTorch. He connects it to ChatGPT and assumes familiarity with his earlier makemore videos.
In briefJeremy Howard's 90-minute practical tour of language models. It covers how they work, GPT-4's capabilities and limits, using the OpenAI API, code writing, data analysis and fine-tuning.
In briefAndrew Ng presents unsupervised feature learning and deep learning, showing how algorithms can learn representations from unlabeled data instead of hand-designed features. He gives examples in vision and audio, including visualizations of features learned by successive layers.
In briefA short infographic video on the distribution of wealth in America, contrasting what people believe, what they consider ideal, and the actual numbers. It highlights how top-heavy US wealth is.
In briefDrew Berry shows scientifically accurate animations of molecular processes inside cells, such as DNA replication, at TEDxSydney.
In briefChris Urmson's SXSW 2016 talk on Google's self-driving car project, covering 1.3 million self-driven miles and the gap between hype and reality. It shows sensor visualizations and playbacks of hard situations the cars met, and discusses failures such as the low-speed bus collision.
In briefHans Rosling and his son Ola run an audience quiz with global population, health and income data to show how wrong most people are about the world. Ola then gives four ways to get less ignorant.
In briefRoy Beck's short gumball presentation using World Bank and Census data to argue that letting people immigrate to the US cannot solve world poverty because the numbers are so large. It is made by NumbersUSA, an immigration-restriction advocacy group.
