KNOWLEDGE NOTESKnowledge Notes
Explore curated knowledge, course notes and methods, with their existing content and sources preserved.
This index lists complete English editions. Additional Chinese notes and bilingual updates remain available in the Chinese index.
350knowledge notesComplete collection of LLM large model interview fundamentals, architecture, and components. Covers three Decoder architectures (prefix/causal/encoder-decoder) + Layer Norm (LN/RMS/Deep Norm) + activation functions (FFN/GeLU/Swish/GLU series) + Attention upgrades (MQA/GQA/Flash/linearized/sparse) + Transformer operations + loss functions (KL/cross-entropy) + similarity functions (cosine/Euclidean/Jaccard) + advanced LLM interview topics (generative models/repetition problem/8 major mitigation strategies) + LLM training experience posts
A synthesis of the MCP protocol ecosystem from 4 perspectives. newtype Lao Huang (evolution prediction from DOS→Win95→Internet explosion + Lao Huang's own Obsidian/Flux/Prompt House MCPs + commonly used MCP list + 7 major configuration pitfalls) / Wang Kai (FMP financial MCP + Chrome MCP modification + Feishu CLI / self-built MCP to solve data accuracy) / Lao D (domestic rap.dev + Xiaohongshu MCP server + Playwright) / AWS industry (Chang Gaowei's ANP protocol comparison with MCP). Covers the MCP vs GUI Agent debate, Smithery's failure, the default Sequential Thinking + Tavily stack, and the closed-loop MCP+n8n workflow.
An image with depth or a video rotating around an object may not provide an asset that can be edited, animated and placed in a game. Hu moves from graphics fundamentals through representations to Meshy’s product and startup choices. The central thread is that outputs need to enter existing workflows to move beyond demonstrations.
Huang Yihe's judgments and practical experience from 2024-04 to 2026-05 on the 5 key characteristics of Agent / Multi-Agent / Workflow / Agentic Workflow / Agent Native / CLI Agent / Multi-Agent Orchestration. Covers the fatal weakness of Manus / limitations of Claude Code Sub-agents / newtype Profile design / 3 rules for Agent delivery rate / three-layer framework for the AI era
Huang Yihe 2025-03 → 2026-05 Full observations on MCP. Basic concepts / MCPs built by Lao Huang himself / List of commonly used MCPs / Configuration troubleshooting / DOS → Win95 evolution prediction / Context engineering vs Prompt engineering / Skill + MCP collaboration / Buffett replication SOP
Huang Yihe's full accumulation on prompt and context engineering from 2024-03 to 2026-05. From the early "shortcut command Prompt" to Prompt House productization, to Karpathy Context Engineering, to the 2026 company-wide rollout of Skill. Covers Thinking-Claude performing thinking / Ultra think 32K thinking budget / structured text-to-image JSON / Market Insights GEM / advanced prompt usage / book summarization Prompt
Huang Yihe 2024-03 → 2026-05 RAG practice. Early AnythingLLM/Open WebUI/Pinecone → Microsoft GraphRAG ($11 per query) → Cohere Command R+ → Milvus + MCP self-built vector database → 2026 Obsidian + NotebookLM final state. Key RAG optimization points / comparison of how different models handle RAG retrieval / document preprocessing / Embedding selection / Lao Huang's personal solution evolution
Huang Yihe 2024-03 → 2026-05 Full evolution of model evaluation/fine-tuning/hardware real-world testing. 505 posts, covering the evolution of GPT/Claude/Gemini/Grok/DeepSeek/Qwen/Llama/Kimi/GLM/Minimax, Mac cluster benchmarks, Unsloth GRPO fine-tuning, Scaling Law debates, output length bottlenecks, TPU vs GPU, the strategy of general base models plus upper-layer vertical fine-tuning, model routing, and model committees.
4 high-quality Prompt templates + 1 CLAUDE.md sample curated by newtype (Huang Yihe). YouTube Podcast Deep Comprehension for Gemini v2 (5-step method: chapter extraction → retelling → bilingual → topic background → learning assistance) / Social Media Deconstruction (5 frameworks: Core Insight / Structure / Psychology / Virality / Takeaways) / Creative Divergence Mode (4 lenses: Gap Analysis / Reframing / Cross-Domain / Breakthrough) / Weekly Synthesis Report (Markdown folder analysis → cross-document synthesis) / CLAUDE.md MCP configuration sample (Tavily + Sequential Thinking + Prompt House + Super Analyst Pro). This is a supplement to the "mature prompt library" of [[newtype-prompt-engineering]]
A complete breakdown of 3 representative works in newtype's Skill system. Super Analyst v1 (intelligent selector for 12 analysis frameworks + integration with prompt-house-local MCP) / Super Analyst Pro 2.0 (7-Stage Workflow + dual Sequential Thinking + Bilingual Web Search + 3-level complexity adaptation) / Super Writer (Material Collection + 7 Style Analysis methods + 6 Writing Methodologies + triple Sequential Thinking). Each Skill is an engineering paradigm of "Sequential Thinking + MCP + framework library". It is the practical proof of the judgment "Skill-ification is the end state" in [[newtype-mcp]]
An AI can retrieve information or fill out forms without knowing which result matters to you. Tao uses this distinction to explain the move from MindOS to Me.bot: first build AI that understands an individual’s needs, then let it call tools providing services. He also wants AI to keep learning from daily experience rather than depend permanently on its original workflows.
A fusion of Prompt Engineering from 4 perspectives. newtype Lao Huang (186 posts + 4 production-grade templates + 5-element task decomposition + Thinking-Claude performative thinking + Ultra think three tiers + Karpathy Context Engineering) / Wang Kai (6000+ words on PUA-ing Claude + using AI as a ruler for quantification + risk-scoring prompts + decision interrogation SOP) / Lao D (three-layer prompt architecture: meta/system/user + Markdown imitation writing SOP + 4-step blogger imitation) / Anthropic internal (Cal Rueb CLAUDE.md anti-patterns). Covers the counter-consensus view on "whether Prompt Engineering needs to be learned", Skill-ification as the end state, Prompt scheduling Prompt, and Sequential Thinking scheduling.
Integration of RAG and knowledge base engineering from 5 perspectives. newtype Lao Huang (5 generations of architecture evolution: AnythingLLM → Cursor+Obsidian → Claude Code AI OS → newtype OS → wiki.newtype.pro / 5 key points of RAG optimization / Milvus+MCP database-model separation / end state: own content → Obsidian, others' content → NotebookLM) / LLM Interview Technical School (14 special topics: R/A/G three modules + PDF parsing + table recognition + text chunking + HYDE + hard negative samples + RAG-Fusion + GraphRAG) / Lao D's counterintuitive teaching ("RAG does not make the model stronger; it automatically strengthens the prompt") / Wang Kai (Agentic Search outperforms RAG + self-built data MCP) / Cal Rueb (Anthropic does not use RAG MCP). Covers the three-way choice among RAG vs Long Context vs NotebookLM, the economics of GraphRAG, and the philosophy of knowledge bases.
Integration of Scaling Law + AGI + Software 3.0 from 4 perspectives. Karpathy's June 2025 talk (Software 1.0/2.0/3.0 triad + LLM = OS/Utility/Fab three analogies + LLM psychology + Iron Man Suit 4 principles + Agentic Decade 2025–2035) / Aschenbrenner's 165-page Situational Awareness (AGI by 2027 + OOMs counting + Trillion-dollar cluster + Intelligence Explosion) / Three-school debate (Zhu Xiaohu application layer / Cohere Aidan pragmatic / Kai-Fu Lee ceiling / Jensen Huang synthesis) / Wang Kai "Viewing the present from an AGI perspective" + Embiricos AGI bottleneck = human typing speed. Covers AGI timeline disagreements, Software 3.0 adoption forecasts, Vibe Coding natural language programming
Wang explains agents through models, environmental feedback and tool-based action. In a product, these form a continuing process: observe the current situation, invoke a tool, obtain a result and use the new information to decide what to do next.
Meng used a mature training platform inside a model company. Starting a business meant building platforms and finding a complementary opportunity. Turning training into a service simplifies data submission and execution without automatically making the data valuable.
Listen: 01:55–06:14. Classifying a cat or dog involves a labeled answer and one decision. Playing a game requires many choices—left, right, or jump—and several paths may win. Reinforcement learning, or RL, improves an action policy using rewards received after acting. It addresses processes with multiple possible choices whose quality becomes clear only later.
RC runs his company on Slock, growing from one person to seven people and forty agents. That describes the contemporary team, not an experimentally optimal ratio. More useful are his responses to duplicated tasks, misplaced context, drifting identity and limited human review capacity.
Qu Kai asks Zhang Hailong, who has more than twenty years of programming experience, to explain the excitement around Cursor. A useful distinction emerges: helping a person operate faster and accepting a task for independent completion are different product promises.
Aschenbrenner's 165-page Situational Awareness manifesto + How People Use ChatGPT (OpenAI) + Why LLMs Hallucinate (OpenAI) + Generative AI Seniority-Biased Change + large-scale Reddit persuasion experiment + OpenAI Practical Guide to Building Agents + Google Prompt Engineering (Lee Boonstra) + Looking at DAO 2025 (Weijin Research) + DeepSeek primer + AGI introduction. 10 documents that together answer five core questions: "What is AGI / How soon will it arrive / How to use it / Who will be impacted / How to build Agents"
