Technology
Artificial Intelligence
We're living through the most significant technological shift since the internet. AI—particularly large language models (LLMs)—has gone from research curiosity to daily tool in just a few years. This page covers what AI actually is, where it came from, the different types, and which models matter right now.
The AI Timeline
Key moments that brought us here.
June 2017
"Attention Is All You Need"
Google researchers publish the Transformer paper—the foundation for every major LLM.
June 2018
GPT-1 Released
OpenAI introduces the first GPT model. The era of generative pre-trained transformers begins.
November 2022
ChatGPT Launches
100 million users in 2 months—the fastest-growing app in history.
March 2023
GPT-4 Released
Multimodal capabilities. Can see images, pass bar exams, write code at near-human level.
March 2024
Claude 3 Opus
Anthropic releases Claude 3. Opus matches or exceeds GPT-4. Real competition emerges.
December 2024
Reasoning Models
o1 and Gemini 2.0 introduce "thinking" modes—step-by-step reasoning before answering.
2025
The Agent Era
AI moves from chat to action. Models browse, code, execute—not just respond.
Hover over dots to explore
Types of AI
Different models for different tasks. Click to explore.
LLMs
Image
Video
Audio
Code
Agents
Large Language Models
Text in, text out. Conversation, writing, analysis, code generation. The core of modern AI.
GPT-4
Claude
Gemini
Llama
Image Generation
Text prompts to images. Art, design, visualization. Quality has become indistinguishable from photography.
DALL-E 3
Midjourney
Stable Diffusion
Flux
Video Generation
Create video from text or images. Still emerging but advancing fast. Hollywood-level quality within reach.
Sora
Runway Gen-3
Veo 3
Kling
Voice & Audio
Text-to-speech, speech-to-text, voice cloning, music generation. Realistic voice synthesis is here.
ElevenLabs
Whisper
Suno
Udio
Coding Assistants
Specialized for code. Autocomplete, generation, debugging, refactoring. 10x developer productivity gains.
Cursor
GitHub Copilot
Codeium
Windsurf
Robotics & Agents
Vision-language-action models. AI that can see, reason, and act in the physical world. The next frontier.
Helix
RT-2
PaLM-E
Understanding AI
Select your level to explore concepts.
Core
Intermediate
Pro
Fundamentals
The basics of how this all works.
- What is an LLM? — A large language model is a neural network trained on massive amounts of text. It learns patterns and can generate human-like text.
- Training — Models are trained on internet text, books, code, and more. The larger the model and data, generally the smarter it gets.
- Tokens — AI thinks in tokens (roughly 4 characters). Context windows measure how much text a model can "see" at once.
- Parameters — The "size" of a model. GPT-4 has ~1.8 trillion parameters. More parameters = more capacity to learn patterns.
- Prompting — How you ask matters. Clear, specific prompts get better results. This is an actual skill to develop.
- Hallucinations — AI can confidently say wrong things. Always verify important information.
Deeper Understanding
Concepts for getting more from AI.
- Fine-tuning — Training a model further on specific data for specialized tasks. Makes general models domain-specific.
- RAG (Retrieval Augmented Generation) — Connecting AI to external knowledge bases so it can cite real, current information.
- Context Windows — GPT-4 handles 128K tokens, Claude 3 handles 200K, Gemini 1.5 handles 1M+. Bigger = more context.
- Multimodal — Models that handle multiple types: text, images, audio, video. GPT-4V, Gemini, Claude 3 are all multimodal.
- System Prompts — Hidden instructions that shape model behavior. Apps use these to create personas and set guardrails.
- Temperature — Controls randomness. Low = predictable, high = creative. Different tasks need different settings.
Advanced Concepts
The cutting edge of AI development.
- Reasoning Models — o1, Gemini Deep Think, Claude "thinking". Models that reason step-by-step before answering complex questions.
- Agents — AI that can take actions: browse the web, write and run code, interact with APIs. The next frontier.
- MCP (Model Context Protocol) — Anthropic's standard for connecting AI to external tools. Databases, APIs, browsers, and more.
- Mixture of Experts — Architecture where different "expert" networks handle different types of queries. More efficient at scale.
- Open vs Closed — Llama, Mistral are open-weight (can run locally). GPT, Claude are closed (API only). Trade-offs in control vs capability.
- Inference Optimization — Techniques like quantization and speculative decoding to run models faster and cheaper.
Top Models by Usage (2025)
The most used AI models worldwide based on weekly active users.
ChatGPT (OpenAI)
800M weekly users
Gemini (Google)
350M weekly users
Claude (Anthropic)
100M weekly users
Llama (Meta)
~50M deployments
Grok (xAI)
~30M via X
Note: Usage != quality. Claude is widely considered the best for coding and long-form writing despite lower consumer usage.
Best Coding Models (My Rankings)
What I actually use for development, ranked by effectiveness.
- 1. Claude 3.5 Sonnet — Best overall for coding. Excellent context handling, follows instructions precisely, great at refactoring. My daily driver in Cursor.
- 2. Claude Opus — For complex architecture decisions and debugging gnarly problems. Slower but smarter for hard tasks.
- 3. GPT-4o — Strong all-rounder. Good for variety when Claude hits a wall. Excellent at explaining code.
- 4. Gemini 2.0 Flash — Incredibly fast, huge context window. Great for large file analysis.
- 5. DeepSeek Coder — Surprisingly good open-source option. Runs locally, no API costs.
Key Players
The companies and people shaping AI.
- OpenAI — Started the LLM race with GPT. Sam Altman (CEO), created ChatGPT. $150B+ valuation.
- Anthropic — Founded by ex-OpenAI researchers focused on AI safety. Dario Amodei (CEO). Created Claude.
- Google DeepMind — Merged Google Brain + DeepMind. Demis Hassabis (CEO). Created Gemini, AlphaFold.
- Meta AI — Yann LeCun leads. Open-source approach with Llama. Massive research output.
- Ilya Sutskever — OpenAI co-founder, chief scientist. Co-authored the Transformer paper. Left OpenAI in 2024 to start SSI.
- NVIDIA — Makes the GPUs that power everything. Jensen Huang (CEO). $3T+ market cap.
Where We're Headed
AI is moving from "chat" to "action." The next wave is agents that can actually do things—browse, code, operate software, and complete multi-step tasks autonomously. Tools like Cursor are early examples: AI that doesn't just suggest code, but writes, tests, and iterates on entire features.
The models will keep getting smarter, faster, and cheaper. The real question isn't whether AI will transform work—it's how fast you learn to work with it.