How to Choose a GPU for Training and Deploying LLMs
Training and inference need different GPUs. Learn VRAM math, a VRAM-to-capability table, per-scenario picks, rent-vs-buy, and China-market caveats.
llm
How to Choose a Base Model for Chat: A Practical Selection Checklist
Picking the right base model matters more than picking the strongest one. Six criteria, the assistant-tone trap, and a 3-day bake-off.
llm
What Is SFT? A Practical Guide to Supervised Fine-Tuning Your Own Model
SFT is the first and most important step in training your own model. Learn dataset formats, assistant-only loss, and common pitfalls.
llm
How to Train an LLM: The Complete Path from Pretraining to Fine-Tuning
Training an LLM means more than feeding it data. Learn pretraining, SFT, and alignment, plus LoRA, QLoRA, RLHF, DPO, and distillation.
llm
Wealth Is Cultivation: Which Realm Are You On?
Why do executives, founders, and capitalists earn more — is it really because they work harder? Chinese cultivation fantasy spelled out the answer long ago: your wealth ceiling is roughly the scale of the rules you can influence.
Thinking
Every Task You Did Today — Who Was It Really For?
Standup, feature work, production fires, interviews… you were busy all day, yet can't say who it was for. Sort each task by who it benefits, and you'll see where your time actually goes.
思考
Couple Trip Planning: I'm Done Being the Middleman Between AI and My Girlfriend
Two people planning a trip, but each talking to AI separately and relaying answers back and forth. There's a better way — just add everyone to one group chat.
vicvic.im
GPT vs Claude vs DeepSeek: One Group Chat to Compare Them All
Worried a single model might be wrong? Create a group chat in vicvic.im, add Claude, GPT, and DeepSeek, ask one question, and compare all answers side by side.
vicvic.im
Claude Code Agent Loop: Dissecting the Heart of an AI Coding Assistant
How does Claude Code understand your requests, invoke tools, and self-recover step by step? A source-code deep dive into the Agent Loop's core architecture — streaming responses, parallel tool execution, auto-compaction, and error recovery.
Claude
Claude Code settings.json Hooks: Auto-Run Scripts at Every Step
Want to auto-validate commands before they run? Send Slack notifications when tasks finish? Claude Code hooks let you inject custom scripts at key moments — PreToolUse, PostToolUse, Stop, Notification — with practical examples.
Claude
Where Is Claude Code settings.json? 5 Config Files, 1 Priority Rule
Claude Code has 5 config file locations: user-level (~/.claude/), project-level (.claude/), local-only, CLI flags, and enterprise managed settings. Learn which file to use when, how priority works, and why your settings might be ignored.
Claude
Claude Code settings.json: Copy-Paste Templates for env, model, auth & More
Ready-to-use Claude Code settings.json snippets for env vars, model switching, apiKeyHelper auth, git signing, effort level, language, auto-updates, and more. Every field includes a working JSON example you can copy and paste.
Claude