LEARN
AI Learn
Learn AI from scratch — what RAG, agents, prompts, fine-tuning, alignment, and context windows actually mean; how to pick a tool; practical use cases; advanced techniques.
What does 'open source' actually mean for an LLM?
Most 'open source' LLMs aren't fully open. They're 'open weights' — you can download and run them, but the training data, recipes, and licenses are messier than the term suggests.
How to pick the right LLM for your use case in 2026
Claude, GPT-5, Gemini, DeepSeek, and Llama all overlap. The right pick depends on what task, what budget, and how much control you need — not which lab has the loudest PR.
How to build your first RAG stack in 2026: a practical pick guide
Embedding model, vector DB, retriever, reranker, generator — five layers, dozens of options. Here's the no-overengineering path.
How to pick a vector database: pgvector vs Qdrant vs Pinecone vs Weaviate
Most teams pick the wrong vector DB for the wrong reasons. Here's how to actually choose based on your stack.
Tokens vs words: how LLM pricing actually works
Why a Chinese paragraph costs more than the same English one, and how to estimate your bill before you ship.
Why input tokens cost less than output tokens
Reading is fast, writing is slow — the technical reason your LLM bill looks the way it does.
Quantization for LLMs: why a 4-bit model still works
How a 70B model gets squeezed onto a single consumer GPU — and where the magic stops working.
30 AI terms every beginner should know in 2026
Skip the academic textbook. Here are the 30 terms — LLM, RAG, embedding, MoE, agent, fine-tuning — defined in one sentence each, with the context that actually matters.
Which LLM is best for Chinese-language tasks in 2026?
Frontier closed models handle Chinese decently, but open-weight Chinese models like Qwen 3 and DeepSeek often win on native tone and cost. Here's how to pick by task.
How to evaluate your RAG system honestly
Vibe-checking 10 queries isn't evaluation. Here's the framework that catches the bugs your demo can't.
How to pick an AI meeting notes tool: Otter vs Fireflies vs Fathom vs Granola
Different tools win different workflows: solo founder, sales team, recurring stand-ups, customer interviews. Match them up.
Open-source LLM vs frontier API: which one for which task in 2026
Open-source models closed most of the gap on commodity tasks. They didn't close it on the hard ones. Here's the line.
How to self-host an LLM stack on a single GPU box in 2026
vLLM, Ollama, LM Studio, LocalAI — pick the right tool for whether this is a hobby, a side project, or a production workload.
Free vs paid AI tools: when does paying actually pay off
Most people overpay or underpay. Here's a clear framework for which AI subscriptions actually earn their keep.
Perplexity vs Felo vs Metaso: AI search engine pick guide for 2026
AI search has split into Western and Chinese-led options. Pick the one that fits your sources and your language.
Build a personal RAG over your notes in an afternoon
Stop scrolling through Notion looking for that one quote. Here's the simplest stack that actually works on personal notes.
Automate customer service Tier-1 with an LLM (without making it worse)
Most LLM customer service deployments fail. Here's the scoped, honest version that actually deflects tickets without enraging users.
Translate a blog into 3 languages with LLM + spot-check workflow
Don't run blog posts through Google Translate. Here's the workflow that produces translations readers don't notice are translations.
Summarize research papers without losing nuance
Generic "summarize this paper" prompts strip the things that matter. Here's how to ask for the parts you actually need.
Draft a weekly newsletter with AI without sounding like a robot
AI-drafted newsletters fail when they read like AI. Here's the workflow that keeps your voice while saving the time.
Use AI to write a resume that gets past the keyword filter
Most resumes are filtered by software before a human ever sees them. Here's how to use AI to fix that without sounding fake.
Use AI to prep for technical interviews
Solo mock interviews, behavioral question rehearsal, and topic gap-finding — AI is unusually good at the parts that intimidate candidates.
Run customer research interviews with AI synthesis
AI doesn't replace the interviewer. It dramatically speeds up everything before and after — from prep to synthesis.
Use AI for UX copy: when it works, when it ruins your tone
AI is fast at draft buttons and tooltips. It's bad at the parts that define how your product feels.