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 notesSynthesis of LLM fine-tuning and alignment. llm-interview-finetune (19 PDFs / fine-tuning basics + PEFT + LoRA family + QLoRA + continued pretraining + RLHF + PPO + DPO + VRAM optimization + inference + Prompting + SFT data) / newtype fine-tuning practice + Wang Kai's LoRA application notes. Covers full-parameter vs PEFT, LoRA + QLoRA, RLHF vs DPO, SFT data construction, and VRAM optimization strategies
Comprehensive overview of LLM core fundamentals. Covers llm-interview-basics (three Decoder architectures: prefix/causal/encoder-decoder + Layer Norm LN/RMS/Deep + activation functions GeLU/SwiGLU + Attention upgrades MQA/GQA/Flash + Transformer operations + loss/similarity functions + repetition problem + 8 major mitigation strategies)/ Karpathy nanochat (reproducing LM in 243 lines of code)/ newtype Karpathy (Software 2.0 neural network weights). Covers three Transformer paradigms, mainstream LLM configurations (LLaMA/GPT/GLM), Attention evolution, and training experience
Complete illustrated 8-PDF set for LLM large model distributed training interviews. Covers distributed training for large models (overview) + Pipeline Parallelism + nn.DataParallel + nn.parallel.DistributedDataParallel + torch.multiprocessing + AMP mixed precision training + DeepSpeed + accelerate + ZeRO learning. Systematically covers the three mainstream distributed training stacks: PyTorch, DeepSpeed, and Megatron
15 special-topic PDFs for the fine-tuning / RLHF / PEFT / continued pre-training section in LLM interviews. Covering LLM fine-tuning interview questions + PEFT (Parameter-Efficient Fine-Tuning) + Adapter-tuning + LoRA series + QLoRA + PEFT library + continued pre-training + continued pre-training sample concatenation + LoRA secondary pre-training + evaluation interview questions + reinforcement learning interview questions + RLHF and its variants + PPO + NLP applications of reinforcement learning + training set interview questions + LLM-generated SFT data + GPU memory optimization + inference interview questions + Prompting section
A full collection of 14 special-topic PDFs on RAG (Retrieval-Augmented Generation) for LLM interviews. Covers RAG fundamentals (why an external knowledge base is needed / the R/A/G three modules) + LangChain RAG hands-on practice + large model external knowledge bases + RAG experience-based interview questions + PDF parsing + table recognition + text chunking + large model-assisted recall (HYDE) + hard negative sample mining + RAG evaluation + RAG optimization strategies + key pain points of RAG + RAG-Fusion + Graph RAG + 8 optimizations for long-term memory in multi-turn conversations
Complete 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.
Listen: 01:57–06:39 and 18:48–19:43. Yu lists doctors, lawyers, engineers specializing in particular languages, and people combining unusual disciplines or languages. A model company may know something is wrong without knowing how to evaluate medical or legal answers systematically. It needs judgment as well as labor.
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.
Monica’s answer has three layers: appear where people work, connect consecutive steps, and use revenue and organizational capability to sustain improvement. A browser extension provides entry into the work itself. Buying an extension is a business tool for accelerating validation.
Integrated needs analysis from 3 perspectives. yitang Truman (4-character formula for needs analysis: Decompose-Deduce-Evaluate-Calculate + 6 proficiency levels + JTBD one-pager L1-L4 + AI collaboration version) / yitang Product Core (4 keywords: User/Willing to Choose/Minimum/Solution + 3 principles + L3-L6 proficiency levels + Product Core Canvas) / Wang Huiwen Tsinghua Product Course (4 key traits of needs + supply-demand relationship + 4P + STP) / wiki-ai-pm-interview (user needs + data closed loop). Covers JTBD framework, micro-sense cultivation, and Top-Down/Bottom-Up ceiling calculation.
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 2024-09 → 2026-05 AI Coding tool stack practice. Cursor-dominant phase → v0+Claude+Cursor three-piece suite → Cline self-built MCP → Gemini 2.5 large-context codebase scanning → Claude Code main battlefield → Codex/GitHub Copilot/OpenCode + newtype-os multi-Agent orchestration. Vibe Coding experience, AI Coding division of labor philosophy, Cursor internal modes, Ultra think, Plan Mode, Claude Code Sub-agents
Huang Yihe's AI tool stack usage and recommendation evolution from 2024-03 to 2026-05. Local model layer (Ollama/LM Studio/MLX) / Frontend layer (AnythingLLM/Open WebUI/LobeChat/ChatWise/Cherry Studio) / Knowledge base layer (MaxKB/RAGFlow/Obsidian Copilot) / Mobile (Enchanted/Apollo) / Configuration and performance (Mac mini M4 / 3060 / DeepSeek-R1 performance comparison) / Lao Huang's daily combo (Gemini + Grok + Raycast + Cursor + NotebookLM)
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's non-technical indie development roadmap from 2024-09 to 2026. Blog → Prompt House → LearnFromYouTube → Super Analyst Skill → newtype Profile/CLI/OS. Tool stack (Next.js + Supabase + Stripe + Cursor + Electron) / Business model (global merchant + tax filing) / Network setup (VPS+VLESS+Reality+SOCKS5) / Go Fundamental design philosophy / AI Coding division of labor
Huang Yihe 2024-03 → 2026-05 Complete collection of AI business judgments, industry observations, and investment reflections. 452 posts, covering the commercial wars among OpenAI/Anthropic/xAI/Apple, the three-faction debate over Scaling Law between Zhu Xiaohu, Kai-Fu Lee and Cohere, the 5 levels of AI, LLM=OS, Page Rank→Agent Rank, the end of SaaS intermediary monopolies, the 10-year logic of NVDA accelerated computing, AI-era startup methodology (Context Matters / Do not build a house on someone else's foundation / Pro Users use multiple tools), personal business experiences (US company / Apple Developer / Substack subscriptions), the three-layer framework (technology stack / value chain / competitive landscape), and three things to do in 2026 (overseas bank card + US stocks + full AI access).
6 essays / reports / philosophical reflections collected by newtype. In-depth analysis of the possibility of an Intel-TSMC joint venture (supply chain / technology sharing barriers / investment outlook INTC vs TSMC) / Trinity Visual Learning (Ex Nihilo Visual Somatic Learning: Jewish Rabbi Tree of Life + Chevra mutual learning resonance + flow economy) / Black Myth: Wukong worldview background (Heavenly Court corruption + Sun Wukong + demon brother alliance) / Awareness and the "I": core paradox and first principles (analysis of the dead end in spiritual epistemology) / Dan Koe startup story (7 failures → $4.2 million annual revenue / compound interest does not arrive when you expect it) / Weekly Report 2026-01-26 to 31 (Lao Yu Indigo dialogue on deep ocean currents + Potemkin AI disrupting software + Founder Park memory asset moat + a16z supply precedes demand)
