AI vs ML vs DL vs LLM: Zenn primer maps how they nest

The primer traces a line from early AI, where rule-based systems like Cyc needed staff to manually enter common-sense knowledge, to machine learning, where models learn patterns from data instead. It then explains that traditional ML still relies on people to define features, which is why deep learning's ability to learn representations from raw data marked a shift in how much of the work the model handles.

The summary also distinguishes open-weight models, whose parameter lists are published and can run locally, from closed-weight models like Claude that are only accessible through apps or APIs. It notes that running a local LLM requires loading all parameters into memory, which is why large-memory Mac configurations saw shortages during a local-AI boom, and that splitting a model across multiple GPUs or offloading parts to disk are options with clear trade-offs in cost and speed.