Gerald Tesoro’s TD-Gammon (early 1990s, IBM) proved that reinforcement learning could reach world-class backgammon by learning from self‑play alone. A small neural network used temporal-difference learning to bootstrap its way toward better play, training on roughly 1.5 million self‑played games with a 3-layer architecture (198 inputs, ~80–160 hidden units, 4 outputs predicting White/Black win with or without a gammon). It barely lost to top players and, in doing so, shifted human strategy (n...

An accessible, concise tour of backpropagation: how the forward pass computes outputs, how the backward pass uses the chain rule to compute gradients efficiently, and why caching intermediates matters. A quick history from 1960s-70s precursors to Werbos, Rumelhart–Hinton–Williams' 1986 breakthrough, with NETtalk and TD-Gammon as milestones. We also discuss limitations like local minima and vanishing/exploding gradients, and what these mean for today’s huge models. Brought to you by Embersilk....

In this quick Deep Dive, we explore why the trefoil is the gateway knot in math: what nontrivial means, how tricolorability proves it’s truly knotted, and why its chirality matters. We also peek at where this elegant three-crossing shape appears in art, symbolism, and biology—from Celtic knots to DNA and protein folding. Note: This podcast was AI-generated, and sometimes AI can make mistakes. Please double-check any critical information. Sponsored by Embersilk LLC

We dive into the strange world of Heterocephalus glaber, the naked mole rat. Discover how a thermoconforming mammal survives extreme underground life, resists cancer with a turbocharged contact inhibition and oversized hyaluronan, and even fuels the brain with fructose under hypoxia. Then explore a colony-level life: a single queen, sterile workers, dispersers, and a social toolkit that looks more like insect society than a mammal group. Finally, we discuss what these radical traits mean for ...

Join us for a deep dive into nanofluids—fluids engineered with nanometer-scale particles to boost heat transfer and beyond. We cover how they're made (one-step and two-step methods), what governs their properties, and three groundbreaking applications: magnetically tunable smart cooling, ultra-sensitive optical sensing, and nanoelectrofuel flow batteries that could transform energy storage. We'll also discuss stability challenges, the subtle science behind the early hype, and a future where r...

We unpack Monash University’s coin-sized, liquid-based neuromorphic chip built from a metal‑organic framework. By guiding ions through nanofluidic channels, it behaves as a nanofluidic memristor and exhibits unprecedented triode‑like nonlinear proton transport. With a hierarchical MOF design that differentiates ions by size, the device offers short‑term memory and a new path toward brain‑inspired computing beyond silicon. Based on the Science Advances paper, we explore what this means for dat...

We trace the spark that launched the modern electrical age—from Arago’s copper-disk curiosities and Faraday’s induction to Bailey’s early motor, Ferraris and Tesla’s AC breakthroughs, and the birth of the three-phase rotating magnetic field. Learn why three phases make a smooth, constant-magnitude field, how slip powers torque, and how this invisible engine drives generators, motors, and the global grid. Note: This podcast was AI-generated, and sometimes AI can make mistakes. Please double-c...

We unpack the anatomy and extraordinary predator toolkit of South America’s terror birds—the flightless, axe-beaked giants that stood up to 10 feet tall, weighed as much as 350 kg, and ruled for tens of millions of years. From bifurcate neural spines and a flexible neck to a dromaeosaur-like sickle claw and functionally didactyl feet, these hunters could deliver rapid, devastating strikes. One genus even migrated into North America during the Great American Interchange. We’ll trace their clim...

Debunk the ‘zero gravity’ myth and explain why astronauts feel weightless because they’re in constant free fall. We explore microgravity, how it’s simulated on Earth (parabolic flights, the vomit comet, and NASA’s drop tower), and the human costs of long-term spaceflight—space adaptation syndrome, muscle atrophy, bone loss, fluid shifts, and orthostatic intolerance. We also look at space-to-earth applications, like growing higher-quality protein crystals for drugs such as pembrolizumab, which...

A deep dive into Mixture of Experts (MoE): how sparse routing selects a tiny subset of experts for each input, enabling trillion-parameter models to run efficiently. We trace the idea from early Metapi networks to modern neural sparsity, explore load-balancing tricks, and see how MoE powers NLP, vision, and diffusion models. A practical guide to why selective computation is reshaping scalable AI. Note: This podcast was AI-generated, and sometimes AI can make mistakes. Please double-check any...

We explore the magnetic flux tube—a self-contained bundle of field lines that channels energy and matter across vast scales. From sunspots and coronal loops to the Io–Jupiter flux rope, and down to the tiny tubes that bind quarks inside protons, this concept sits at the heart of how nature concentrates power and organizes matter. We’ll unpack the flux-conservation idea, the formation of twisted flux ropes, and what this universal structure hints about unity in physics—plus the intriguing ques...

We dive into how limbless movers—snakes, slugs, and more—convert wiggling waves into straight-line propulsion through directional friction. Learn about frictional anisotropy, the isotropic-sleeve experiment that stops forward motion, and the idea of a mechanical rectifier that turns oscillation into sustained thrust. From tribology and deterministic surface textures to biomimetic designs inspired by reptile skin, we explore how nature’s friction tricks could inform more efficient engines and ...

Entrenar un modelo es solo el comienzo. En este episodio te contamos cómo construir pipelines de Machine Learning y llevarlos al siguiente nivel con MLOps, la práctica que conecta la ciencia de datos con la ingeniería para automatizar, desplegar y monitorear modelos en producción.Aprenderás qué herramientas se usan (Airflow, MLflow, Docker, Kubernetes), cómo mantener modelos actualizados y cómo evitar el caos de los “notebooks eternos”.Si quieres que tus modelos no se queden en el laboratorio, sino

In the deep twilight, velvet belly lantern sharks light up with a two-step system—hormones set a stealth glow for camouflage, while neural signals push quick flashes for signaling. We unpack the chemistry (melatonin, prolactin, GABA, nitric oxide), reveal their krill-dominated diet and ontogenetic shifts, and explain how regional environmental baselines shape isotopic niches, showing a level of deep-sea complexity that rewrites what we expect from lantern sharks. Note: This podcast was AI-gen...

We explore Sednoids—detached, highly elongated trans-Neptunian objects whose orbits sit far beyond Neptune. We unpack why their orbital orientations resist easy explanations, review the leading ideas for their origin (stellar flybys, interstellar capture, or a hidden Planet Nine), and discuss how the four confirmed members—Sedna, 2012 VP113, Lele Ekohonoa, and 2023 KQ14—might preserve a fossil record of the Sun's birth environment and the outer solar system's mass. Note: This podcast was AI-g...

We unpack Reasoning Bank, a memory framework that turns execution logs into transferable, strategic knowledge. Learn how failure data becomes counterfactual lessons, how structured memory (title, description, content) captures generalizable reasoning, and how memory-guided 'maps' enable deep, scalable exploration. Explore how emergent, adaptive strategies hint at deeper general intelligence. Brought to you by ember silk.com. Note: This podcast was AI-generated, and sometimes AI can make mista...

When you follow the news, particularly in countries like the UK and the US, it sometimes feels like people are less optimistic about their lives than they were in the past.But a new piece of analysis from polling company Gallup suggests this might just be the local view, not the global one. Using data from the Gallup World Poll, it suggests that “people in more countries are living better lives and expressing more hope for the future” than at any point in the last decade.Tim Harford speaks to

Von Neumann, entre stratégies et circuits

Number of the Day - "382"This podcast is produced by Klassic Studios. Check out our other podcasts there! Learn more about your ad choices. Visit megaphone.fm/adchoices

Chapter 1 1 minute 25 seconds 諜戰劇、電報、密碼Chapter 2 3 minutes 6 seconds 編碼 vs 密碼 Chapter 3 3 minutes 48 seconds 中國電報教父盛宣懷 Chapter 4 4 minutes 34 seconds 作為編碼的語言 Chapter 5 6 minutes 52 seconds 《觸電的帝國》 Chapter 6 7 minutes 23 seconds 電報密碼 Chapter 7 10 minutes 25 seconds 一次一密 Chapter 8 11 minutes 25 seconds 豪密 Chapter 9 13 minutes 48 seconds 日本破譯大清密碼 Chapter 10 15 minutes 32 seconds Enigma Chapter 11 15 minutes 59 seconds 重複是破譯的先兆 Chapter 12 16 minutes 42 seconds