Deep Learning: Stacking Simple Things Until They Get Smart
A neural network is layers of "multiply, add, squash." Backpropagation figures out which numbers to change. Transformers add one trick: attention.
Articles
14 published · newest first
all Machine LearningDeep LearningModelsGPUsPrompt EngineeringContext EngineeringHarness EngineeringLoop EngineeringSemantic LayerChatbots & AgentsModel SwitchingMCP ServersWorkflowsObservability
Machine Learning in One Sitting
The whole field compresses to a loop: guess, measure error, adjust, repeat. Everything else is engineering around that loop.