Examples

The video explains why the DeepSeek models (notably R1/V3) caused such a stir: they show that performance and training/inference costs can be improved without simply "scaling to infinity." The speaker puts this in the context of LLMs (Transformers, next-word prediction), the race for model size, and the "open" vs "closed" choices.

A pedagogical explanation of modern image generators: how a model goes from a "noisy" state to a coherent image, and how text guides the generation. The video mainly serves to clarify the principle (without requiring a huge math background).

The video starts from "classic" (symmetric) encryption and shows its big practical problem: how to share the key securely. It then introduces asymmetric encryption with key pairs (public/private), and the idea of publishing your public key so anyone can encrypt a message intended for you.

A very visual introduction to neural networks, built around a concrete goal: recognizing handwritten digits. The video establishes the intuition of neurons, layers, and how a network represents a function that maps pixels to a prediction.

Starting from an "absurd" dance sequence, Derek Sivers shows that the real start of a movement depends less on the "leader" than on the first follower who makes the action legitimate and imitable. Then, the movement grows by making participation simple and visible.

The video offers simple reflexes to avoid spreading misinformation: verify via fact-checking sites, compare multiple reliable sources, and slow down before sharing (especially when content triggers an emotional reaction).

An animation that contrasts "linear" economy (extract - make - dispose) with a "restorative" circular economy: rethinking product design so they can be repaired, reused, reconditioned, and powered by renewable energy.