camelbell
AI Superpowers: China, Silicon Valley, and the New World Order
Loading…

AI Superpowers: China, Silicon Valley, and the New World Order

Kai-Fu Lee

4.51113 learners

From Kai-Fu Lee’s 40 years at the forefront of AI research and industry, *AI·Future* unpacks the revolution reshaping our world. Starting with AlphaGo’s historic 2017 win over Go champion Ke Jie—China’s 'Sputnik moment' for AI—the book breaks down the divergent paths of US and Chinese AI development: America leads in foundational research and top talent, while China leverages massive real-world data and aggressive government policies to dominate AI applications. Lee outlines four waves of AI evolution, from internet recommendation engines to autonomous robots, and warns of urgent challenges: job displacement hitting white-collar workers first (per Moravec’s Paradox), widening wealth gaps, and overhyped fears of superintelligence. Drawing on his battle with lymphoma, Lee reframes the conversation: AI can’t replicate human empathy, love, or creativity. He proposes solutions like social contribution stipends to build a human-centric future where AI serves as a tool, not a threat, and we prioritize connection over efficiency.

Artificial IntelligenceUniversal Basic IncomeDeep LearningUS-China AI RivalryMoravec’s Paradox
Deep Learning
Loading…

Deep Learning

Ian Goodfellow；Yoshua Bengio；Aaron Courville

4.6775 learners

Written by leading AI experts Ian Goodfellow, Yoshua Bengio, and Aaron Courville, *Deep Learning* is the definitive textbook for mastering this transformative subfield of artificial intelligence. Spanning from foundational mathematical tools (linear algebra, probability, numerical optimization) to cutting-edge generative models, it caters to both students building a rigorous theoretical knowledge base and engineers seeking practical implementation guidance. The book systematically covers core topics: machine learning fundamentals, deep feedforward networks, optimization challenges unique to neural networks, specialized architectures like CNNs and RNNs, real-world deployment strategies, unsupervised learning techniques, and advanced generative models such as VAEs and GANs. It also traces deep learning’s three historical waves, contextualizing key breakthroughs like backpropagation and AlexNet. With clear explanations, mathematical rigor, and actionable insights, this guide equips readers to understand, build, and apply deep learning systems across computer vision, natural language processing, speech recognition, and beyond.

Deep Neural NetworksConvolutional Neural NetworksRecurrent Neural NetworksGenerative Adversarial NetworksBackpropagation
The Thinking Machine
Loading…

The Thinking Machine

Stephen Witt

3.3511 learners

From a bullied Taiwanese immigrant teen in a Kentucky reform school to helming the world’s most valuable semiconductor giant, Jensen Huang’s journey lies at the heart of *The Thinking Machine*. This book chronicles NVIDIA’s improbable rise: starting with the catastrophic failure of its first graphics chip, NV1, Huang and his team pivoted to parallel computing, betting billions on the CUDA platform despite fierce Wall Street skepticism. The 2012 breakthrough of AlexNet—trained on NVIDIA GPUs—unlocked the AI revolution, transforming NVIDIA from a niche gaming hardware maker into the global core of AI infrastructure. It delves into the company’s cutthroat, fast-paced culture, strategic acquisitions like Mellanox, and the fiery debate over AI’s existential risks, contrasting Huang’s optimistic view of AI as a harmless tool with warnings from leading researchers. More than a business memoir, it’s a story of resilience, visionary risk-taking, and how a seemingly narrow tech bet reshaped the future of computing.

Deep LearningAlexNetJensen HuangNVIDIACUDA Platform
Concept

Desirable Difficulty

Short-term frustrating methods (spaced practice, interleaving) help you learn more solidly and flexibly in the long run.

Deep LearningMemoryLearning Method
Range · Chapter 2
Method

Feynman Technique Steps

1. Write the concept you want to understand. 2. Explain it to a child (simple language). 3. Identify gaps and go back to learn. 4. Refine until you can teach it clearly—this exposes weak spots.

Deep LearningFeynman TechniqueUnderstanding
Ultralearning · Chapter 5
Scenario

Google's Cat Recognition

Google's 9-layer deep network learned to recognize cats from unlabeled YouTube videos—no human told it what a cat is!

Deep LearningGoogleUnsupervised Learning
The Master Algorithm · Chapter 4
Scenario

AlexNet Breakthrough In 2012

Krizhevsky trained AlexNet on two GPUs for two weeks. It won the ImageNet competition with error rate half of the second place, ending the long AI winter and bringing deep learning back to mainstream.

Deep LearningAI History
The Alignment Problem · Chapter 1
Concept

Backpropagation Algorithm

Backpropagation adjusts neural network weights by calculating error from output to input. It's how deep learning models learn from data.

Deep LearningNeural NetworksBackprop
The Master Algorithm · Chapter 4
Scenario

DeepMind AI Breakthrough

Google's DeepMind AI used deep reinforcement learning, spent only a few hours to beat human players in half of old arcade games, and even independently discovered hidden bugs and loopholes in the games.

AIDeep LearningTech Innovation
The Inevitable · Chapter 2
Concept

What is Deep Learning

Deep learning is a core AI branch that lets computers learn complex patterns from data via multi-layer feature representations, no heavy manual feature engineering is required.

AIDeep LearningFeature Learning
Deep Learning · Chapter 1
Concept

ReLU Activation Function

Short for Rectified Linear Unit, its formula is g(z) = max(0,z). It keeps gradient non-zero when input is positive, avoiding the gradient vanishing problem common in old activations.

Deep LearningActivation FunctionReLU
Deep Learning · Chapter 4
Concept

Generative Adversarial Network

GAN is a modern generative model that trains two networks (generator and discriminator) to compete, until the generator can produce samples nearly identical to real data.

Deep LearningGANGenerative Model
Deep Learning · Chapter 9
Insight

3 DL Development Waves

Deep learning’s 3 development waves all succeeded or failed due to matching tech conditions. The third wave starting 2006 succeeded thanks to pre-training methods, GPU power and massive public datasets.

tech developmentDL HistoryAI Trend
Deep Learning · Chapter 1
Insight

DL Optimization Difference

Deep learning optimization does not aim for absolute minimum training loss, but for maximum generalization on unseen data. It often adds regularization terms to avoid overfitting training noise.

OptimizationGeneralizationDL Training
Deep Learning · Chapter 6
Insight

Distributed Training Advantage

Distributed training is the only feasible way to train super large deep learning models. Splitting model or data across multiple machines lets teams scale up training speed for big datasets.

ScalingDL EngineeringDistributed Training
Deep Learning · Chapter 6
Insight

Distributed Representation Gain

Distributed representation can represent far more distinct concepts with way fewer parameters than traditional symbol representation. It brings much stronger generalization ability for the trained model.

Deep LearningRepresentation LearningUnsupervised
Deep Learning · Chapter 7
Scenario

2012 AlexNet Breakthrough

Alex Krizhevsky’s AlexNet CNN won 2012 ImageNet contest, its error rate was 10% lower than the second place, marking the full rise of deep learning’s third wave.

AlexNetComputer VisionDL History
Deep Learning · Chapter 1