Course Overview
AI Beginner Crash Course
Cut through the buzzword confusion. Understand what all these names, tools, and concepts actually are — and how they relate to each other.
Scroll
01
Who Builds the Models
The companies at the top of the pyramid. They build the engines — everything else is built on top.
Frontier LLM Companies
Specialist Companies (Image, Video, Audio)
02
What AI Can Do
The types of content AI can work with. "Multimodal" means it handles more than one.
Text
Including code
Image
Photos, art, diagrams
Video
Generated clips & edits
Audio
Speech, music, sound
Input vs. Output: "Multimodal" doesn't tell you the full picture. A model might accept images
as input but only output text. Always ask: what goes in, and what comes out?
03
The Models
The actual AI. The brain. Everything else is an interface sitting on top.
Model
Catch-all term for any AI brain.
LLM
Large Language Model. Text-focused.
Diffusion Model
Architecture behind image & video generation.
Foundation Model
Umbrella term for any large pre-trained model.
SLM
Small Language Model. Lighter, fewer resources.
Open-Weight
Model weights are public. You can download & run it yourself.
Closed
Only accessible through the company's apps or API.
Knowledge Cutoff
Models are frozen in time. They only know data up to their training date.
Company → Model Map
| Company | Model | Input | Output |
|---|---|---|---|
| Anthropic | Claude Opus 4.5 | TextImagesPDFs | Text |
| Claude Sonnet 4.5 | TextImagesPDFs | Text | |
| Claude Haiku 4.5 | TextImagesPDFs | Text | |
| OpenAI | GPT-5.2 | TextImagesAudio | TextAudio |
| GPT-5.2-Codex | TextImages | Text | |
| DALL-E 3 | Text | Images | |
| Sora 2 | TextImages | Video | |
| Gemini 3 Pro | TextImagesAudioVideo | Text | |
| Gemini 3 Flash | TextImagesAudioVideo | Text | |
| Imagen 3 | Text | Images | |
| Veo 2 | TextImages | Video | |
| xAI | Grok 4.1 | TextImages | Text |
| Aurora | Text | Images | |
| Meta | Llama 4 Scout | TextImages | Text |
| Llama 4 Maverick | TextImages | Text | |
| DeepSeek | DeepSeek-V3.2 | Text | Text |
| DeepSeek-R1 | Text | Text | |
| Alibaba Cloud | Qwen3-235B | Text | Text |
| Qwen3-Max | Text | Text | |
| Qwen-Image | Text | Images | |
| Mistral | Mistral Large 3 | Text | Text |
| Devstral 2 | Text | Text | |
| Cohere | Command A | Text | Text |
| Command A Vision | TextImages | Text | |
| AI21 Labs | Jamba Large | Text | Text |
| Jamba2 Mini | Text | Text | |
| Stability AI | Stable Diffusion 3.5 | TextImages | Images |
| Stable Video 4D 2.0 | TextImages | Video | |
| Stable Audio 2.5 | Text | Audio | |
| Midjourney | Midjourney V7 | Text | ImagesVideo |
| Runway | Gen-4.5 | TextImages | Video |
| Black Forest Labs | Flux 2 | TextImages | Images |
04
Where AI Runs
Cloud, local, or a mix of both.
Cloud
Model runs on the company's servers. Access via the internet. No special hardware
needed.
Trade-off: your data leaves your machine.
Local
Model runs on your own machine. Requires GPU + RAM. Only works with open-weight
models.
Upside: complete privacy. Nothing leaves your computer.
Hybrid / Edge
Some processing on-device, complex tasks sent to cloud. Apple Intelligence is a key
example.
Balance of privacy, speed, and capability.
05
Chat Interfaces
The apps you talk to. They're wrappers — not the AI itself.
First-Party (made by the model companies)
ChatGPT
OpenAI — wraps GPT, DALL-E, Sora
Claude.ai
Anthropic — wraps Claude models
Gemini
Google — wraps Gemini, Imagen, Veo
Grok
xAI — wraps Grok, Aurora
Meta AI
Meta — wraps Llama. In WhatsApp, Instagram, Facebook.
DeepSeek Chat
DeepSeek — wraps DeepSeek models
Le Chat
Mistral — wraps Mistral models
Qwen Chat
Alibaba Cloud — wraps Qwen models
Third-Party (use other companies' models)
Microsoft Copilot
Wraps GPT. Built into Windows, Edge, Office 365.
Perplexity
Search-focused. Uses OpenAI, Anthropic, and others. No own model.
Poe
By Quora. Switch between Claude, GPT, Llama, Gemini.
Duck.ai
Privacy-focused by DuckDuckGo. Free, no account.
You.com
AI search + chat. Multiple models.
ChatHub
Side-by-side model comparison.
TypingMind
Bring-your-own-API-key interface.
HuggingChat
Open-source chat for open-weight models.
Interface Features (not model features)
Web Search
Interface searches the web, feeds results to the model as context.
Deep Research
Automated multi-search loops. Web search on steroids.
Memory
Interface stores preferences, injects them later. Model has zero memory.
Conversation History
Interface replays past messages to simulate continuity.
System Prompts
Hidden instructions that shape model behavior.
File Uploads
Interface converts files to text the model can read.
Image Generation
Routes to a separate image model (DALL-E, Imagen, etc.).
Voice Mode
Speech-to-text & text-to-speech. Model still processes text.
Artifacts / Canvas
Renders output in interactive views. Model just wrote text.
06
Running AI Locally
Your machine, your models. No internet, no subscriptions, complete privacy.
Where to Get Models
Hugging Face
The "GitHub of AI models." Largest open platform for downloading models, datasets, and
tools. Free.
Ollama
CLI tool that makes local models dead simple.
ollama run llama3
and go.Local Chat Interfaces
LM Studio
Most popular. Desktop app with model browser, chat UI, and local API server.
Jan
Open-source alternative. Lightweight, offline, imports from Hugging Face.
GPT4All
Simplest entry point. One-click downloads, works on modest hardware.
Local Image Generation
Stability Matrix
One-click installer for local image gen tools. Manages ComfyUI, Fooocus from one
dashboard.
Fooocus
Simple as Midjourney but local. Minimal settings, great defaults.
ComfyUI
Node-based workflows. Steep learning curve, maximum control.
AUTOMATIC1111
The original. Web-based UI with massive extension ecosystem.
07
How It All Fits Together
The model is simple. Everything else is layers on top.
Tools / Plugins
Web search, file uploads, code execution, image generation (via separate model),
memory, plugins, MCP, integrations
Interface (the app / wrapper)
ChatGPT, Claude.ai, Gemini, Grok — conversation history, accounts, settings,
system prompts, UI
Model (the engine)
GPT, Claude, Gemini, Grok — text in, text out. No memory. No web. No
files. Stateless.
08
AI Coding Agents
From chat to coworker. They live inside your project — reading, editing, running, iterating.
Browser-Based App Builders
Lovable
Describe an app, get a full-stack React app. Supabase integration. One-click deploy.
$100M ARR in 8 months.
Bolt.new
Open-source. Generate & deploy web apps from prompts. Fastest for prototypes.
Replit
Full cloud IDE with AI agent, database, auth, hosting. 40M+ users.
v0
Vercel's builder. Generates frontend UI from text or image prompts.
Google AI Studio
Free prototype builder + Gemini playground. Firebase Studio for full workspace.
First-Party Coding Agents (built by model companies)
Claude Code
Anthropic. Reads codebase, edits files, runs commands, handles git, creates PRs. Most
access points of any agent.
Codex
OpenAI. Open-source. Cloud mode runs parallel sandboxes. Powered by GPT-5.2-Codex.
Gemini CLI
Google. Open-source. Reason-and-act loop. Generous free tier.
How You Access Them
| Agent | Terminal (CLI) | VS Code | Desktop App | Web Interface |
|---|---|---|---|---|
| Claude Code | ✓ | ✓ | ✓Claude Desktop | ✓claude.ai/code |
| Codex | ✓ | ✓ | ✓Standalone app | ✓chatgpt.com/codex |
| Gemini CLI | ✓ | ✓Companion ext. | — | — |
AI-Native IDEs (third-party wrappers)
Cursor
VS Code fork. Multi-file editing, autonomy slider. Uses OpenAI, Anthropic, Google
models.
Windsurf
VS Code fork. Free tier. Autonomous agent for packages, files, refactoring.
Google Antigravity
VS Code fork. Manager View for orchestrating multiple parallel agents.
IDE Extensions
Claude Code 1st party
Same CLI agent, in VS Code. Reads, edits, runs, handles git.
Codex 1st party
Same CLI agent, in VS Code.
Gemini Code Assist 1st party
VS Code & JetBrains. Agent mode, multi-file edits. Free for individuals.
GitHub Copilot 3rd party
Autocomplete + Agent Mode. Also has a fully autonomous Coding Agent for GitHub issues.
Cline 3rd party
Open-source. Autonomous agent. Supports multiple model providers. Pay per API use.
Cloud Agents (fully autonomous)
09
The Developer Toolkit
Not AI products — the environment you work in. The workbench, the tools, the infrastructure.
Code Editors / IDEs
VS Code
Microsoft. Most popular editor in the world. Free, open-source. What Cursor, Windsurf,
Antigravity fork from.
JetBrains
IntelliJ, WebStorm, PyCharm. Professional-grade, language-specific. Paid.
Zed
Written in Rust. Extremely fast. Has AI features. Growing quickly.
Sublime Text
Fast, lightweight, minimal. Loyal following.
Vim / Neovim
Terminal-based, keyboard-driven. Steep curve, extreme power.
Xcode
Apple. Required for iOS / macOS development.
Terminal Apps
Warp
Most AI-forward terminal. Built-in agents, natural language commands. Mac, Linux,
Windows.
Ghostty
Fast, GPU-accelerated. By HashiCorp co-founder. No AI, just speed. Open-source.
Kitty
GPU-accelerated. Tabs, images, splits, SSH. Highly configurable.
Alacritty
Fastest terminal. Rust, GPU-accelerated, intentionally minimal.
WezTerm
Cross-platform, GPU-accelerated. Lua scripting for deep customization.
Windows Terminal
Microsoft. PowerShell, WSL, Git Bash in one app. Standard for Windows.
iTerm2
macOS. Split panes, session restore. The Mac default for years.
Languages
HTML
CSS
JavaScript
TypeScript
Python
SQL
Rust
Go
Frameworks
React
Next.js
Vue
Svelte
Astro
Django
Flask
FastAPI
Node.js
Express
Other Tools You'll See
10
Git & GitHub
Version control and the connective tissue of modern software development.
Git
Version control system. Tracks every change, manages branches, handles merges. Free,
open-source, on every developer's machine.
GitHub
Microsoft. Dominant platform for hosting repos. Pull requests, issues, automation, free
hosting. 100M+ developers.
GitHub Desktop
Visual git app. Manage repos without the command line. Good for beginners.
11
Database & Backend Services
Where your app stores data. Backend-as-a-Service platforms that bundle database, auth, storage, and APIs.
Supabase
The vibe coding default. Open-source, PostgreSQL. Database, auth, storage, APIs.
Integrated with Lovable & Bolt.
Default
Firebase
Google. The original BaaS. NoSQL (Firestore). Great for real-time and mobile.
Neon
Serverless PostgreSQL. Scales to zero. Just a database — no auth, no extras.
PocketBase
Single-file self-hosted backend. Written in Go. Fully local, no vendor lock-in.
Appwrite
Open-source, self-hostable BaaS. Similar to Supabase but NoSQL-style.
Convex
Real-time backend. Database, server functions, and real-time sync in one.
Xano
No-code backend with visual API builder. PostgreSQL underneath.
12
APIs & API Keys
How software talks to software.
API
Application Programming Interface. How your app talks to external services —
Claude, Supabase, Stripe, anything. Think of it as a waiter: your app tells the API what it wants, the API
goes to the kitchen (the service), and brings back the response.
API Key
A password that identifies your app and ties usage to your account. You get it
from the service's dashboard, paste it into your project, and your app can talk to that service. The #1
thing that trips up beginners in vibe coding.
13
Hosting & Deployment
Putting it on the internet.
Static Site / Frontend Hosting
Vercel
Default for Next.js/React. Zero-config deploys from GitHub. Free hobby tier.
Netlify
Original JAMstack host. Built-in forms and auth. Free tier, 100GB bandwidth.
Cloudflare Pages
Unlimited free bandwidth. 300+ edge locations. Best value at scale.
GitHub Pages
Free hosting from a GitHub repo. Push HTML/CSS/JS and it's live. Dead simple.
Full-Stack / Backend Hosting
Render
Modern Heroku replacement. Web services, workers, cron jobs, managed DBs.
Railway
One-click DB deploys. Scale-to-zero pricing. Great DX.
Fly.io
Deploy containers globally. More technical, multi-region.
Streamlit Cloud
For Streamlit apps (Python dashboards). Connect GitHub, it's live. Free.
Replit
Build and deploy in the browser. Good for prototypes.
Google Cloud Run
Serverless containers. Pay only when requests come in.
The Big Cloud Providers