BestHub
Sign in
Wu Shixiong's Large Model Academy
Author

Wu Shixiong's Large Model Academy

We continuously share large‑model know‑how, helping you master core skills—LLM, RAG, fine‑tuning, deployment—from zero to job offer, tailored for career‑switchers, autumn recruiters, and those seeking stable large‑model positions.

137
Articles
0
Likes
871
Views
0
Comments
Recent Articles

Latest from Wu Shixiong's Large Model Academy

100 recent articles max
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
Jun 26, 2026 · Artificial Intelligence

Claude Code Hooks: How to Bypass Model Self‑Awareness and Enforce Deterministic Actions

Claude Code hooks are deterministic scripts that run at specific events—such as PreToolUse, PostToolUse, SessionStart, and UserPromptSubmit—to enforce hard constraints, bypass the probabilistic suggestions in CLAUDE.md, and enable interception, feedback, analytics, and automation, with detailed configuration, input/output protocols, and safety considerations.

AI automationClaude CodeConfiguration
0 likes · 25 min read
Claude Code Hooks: How to Bypass Model Self‑Awareness and Enforce Deterministic Actions
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
Jun 24, 2026 · Artificial Intelligence

Why Public QA Datasets Fail for Deep Research Agents—and How to Build Effective Training Data

The article explains that single‑ or two‑hop QA datasets cannot teach Deep Research agents multi‑step reasoning, outlines four mainstream data‑construction methods, describes trajectory sampling with a three‑stage funnel filter, and shares practical guidelines on data volume, difficulty distribution, question types, and common pitfalls.

AI Agent TrainingData ConstructionDeep Research
0 likes · 32 min read
Why Public QA Datasets Fail for Deep Research Agents—and How to Build Effective Training Data
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
Jun 23, 2026 · Artificial Intelligence

When RAG Returns Junk, Why a LLM Can’t Fix It – Building an Agentic RAG

The article examines why traditional single‑step Retrieval‑Augmented Generation fails when retrieved passages are irrelevant, outlines the three fundamental flaws of that pipeline, and presents the Agentic RAG paradigm—turning retrieval into a reusable tool with planning, reflection, and decision loops, illustrated with code, interview scenarios, and practical deployment tips.

AIAgentic RAGLLM
0 likes · 32 min read
When RAG Returns Junk, Why a LLM Can’t Fix It – Building an Agentic RAG
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
Jun 20, 2026 · Artificial Intelligence

How I Burned $15K on Claude Code in a Month and Finally Mastered Skill Writing

After spending nearly $15,000 on Claude Code and Codex in a single month, the author discovered that most of his dozens of skills were never invoked, learned the progressive‑disclosure mechanism, rewrote skill descriptions, added verification steps, organized skills as folders with scripts and hooks, and now knows how to identify and optimize the truly useful skills.

AI AgentsClaude CodeSkill Development
0 likes · 19 min read
How I Burned $15K on Claude Code in a Month and Finally Mastered Skill Writing
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
May 26, 2026 · Artificial Intelligence

Why Anthropic Caps SKILL.md at Under 5K Tokens and How to Structure Yours

The article explains Anthropic's official 5K‑token limit for SKILL.md files, breaks down the three‑level loading architecture, demonstrates progressive disclosure with concrete token calculations, and provides a step‑by‑step refactoring guide that reduces token usage while improving skill accuracy.

AI engineeringAnthropicClaude
0 likes · 16 min read
Why Anthropic Caps SKILL.md at Under 5K Tokens and How to Structure Yours
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
May 13, 2026 · Artificial Intelligence

How to Explain a Jump from 71% to 94% Tool‑Calling Accuracy in a JD Interview

The article walks through a JD interview scenario where a candidate explains how a tool‑calling accuracy metric rose from 71% to 94% by detailing the full SFT data‑engineering pipeline, teacher‑model trajectory generation, quality validation, evaluation methodology, and interview‑ready talking points.

Data EngineeringEvaluationFunction Calling
0 likes · 19 min read
How to Explain a Jump from 71% to 94% Tool‑Calling Accuracy in a JD Interview
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
Apr 30, 2026 · Artificial Intelligence

When Is Claude Code’s Memory Injected into system_prompt? Interview Insight

The article explains that Claude Code loads persisted memory once at REPL startup via _build_system(), inserts it as the 10th segment of system_prompt, enforces a 200‑line limit on MEMORY.md, deliberately avoids side‑effects in get_memory_dir(), and only refreshes the prompt with the /model command.

Claude CodeLLMMemory Injection
0 likes · 11 min read
When Is Claude Code’s Memory Injected into system_prompt? Interview Insight
Wu Shixiong's Large Model Academy
Wu Shixiong's Large Model Academy
Apr 29, 2026 · Interview Experience

ByteDance Interviewer Asks: What Rank r Do You Use for LoRA? I Said 64—He Said I'm Wasting GPU Memory

The article examines a common interview scenario where candidates are asked about LoRA rank selection, outlines two typical mistakes—guessing or staying silent—and presents a three‑step strategy of honest boundary setting, logical derivation, and asking a focused question, illustrating the approach with concrete LoRA calculations and a vLLM case study.

AI engineeringLoRAinterview strategy
0 likes · 13 min read
ByteDance Interviewer Asks: What Rank r Do You Use for LoRA? I Said 64—He Said I'm Wasting GPU Memory