📔 Change logs

Updated: Intent Recognition — replaced the summarized English version with a complete translation aligned to the Chinese article
Updated: User Value and Case Study: Yuque’s Consumer Product Line — polished the English translations and completed the Yuque case reference
Updated: “Understanding LLMs from First Principles” 01: Predicting the Next Token — reworked the prose throughout, merging redundant paragraphs, trimming repeated examples, and turning the closing summary into lists
Updated: 06: The Compute Ledger of Training, RL, and Inference with worked exercise answers and a clearer distinction between RL token counting conventions and the α coefficient

Published: 06: The Compute Ledger of Training, RL, and Inference — using 6ND, lifetime inference volume, and RL’s effective cost coefficient to explain lifecycle compute allocation
Updated: “Understanding LLMs from First Principles” 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, 13, and 14 with learning checks and technical appendices for compression, MoE, reasoning models, and the Attention formula
Published: 14: Commercialization and the Future — from SaaS to outcome as a service
Published: Extra: Model Distillation — pouring big-model behavior into smaller models
Published: 13: AI Native Product Design — making probabilistic systems feel reliable
Published: 12: LLM Engineering — KV cache, inference cost, and deployment systems
Published: 11: Agents — from chatbots to task-execution systems
Published: 10: Tool Use — from saying things to doing things
Published: “Understanding LLMs from First Principles” 06: Scaling Laws and Emergence, 07: Inference and Generation, 08: The Nature of Hallucination, and 09: RAG
Updated: gave the “Understanding LLMs from First Principles” series a full polish pass — tightened mechanism accuracy and technical details, standardized and completed section illustrations, and added series entrance cards to the home page

Published: 01: The First Principle of LLMs: Token Prediction
Published: 02: Token and Embedding — how language becomes numbers a model can process
Published: 03: Transformer and Attention — how models “see” context
Published: 04: Language as Compression of the World — why prediction can become intelligence
Published: 05: Pretraining, Fine-tuning, and Alignment — from continuation machine to assistant
Published: “The Math Behind LLM Pricing” 01: How Inference Works, 02: Inference as Equations, 03: From Latency to Cost, 04: Cracking Open the KV Cache, and 05: From One GPU to a Cluster, with interactive simulators
Updated: added section illustrations across “Understanding LLMs from First Principles” 01–05, and added a “Recently Published” section to the home page