# Aiodic – The Periodic Table of AI > Aiodic is the Periodic Table of AI: 118 AI building blocks, from tokens and transformers to agents and AGI, each explained in plain English, plus a blog on how they fit together. > Generated by All in One SEO Pro v5.0.5.1, this is an llms.txt file, used by LLMs to index the site. ## Sitemaps - [XML Sitemap](https://aiodic.com/sitemap.xml): Contains all public & indexable URLs for this website. ## Posts - [Blog](https://aiodic.com/blog/) - [The Periodic Table of AI: What Aiodic Is and How to Use It](https://aiodic.com/periodic-table-of-ai/): Aiodic is the Periodic Table of AI: 118 building blocks, from tokens to agents, each with a plain-English page. Here's how it works and where to start. - [AI Core Concepts: 11 Building Blocks Explained in Plain English](https://aiodic.com/ai-core-concepts/): Core Concepts on the Periodic Table of AI: 11 building blocks, from neural networks and tokens to transformers, explained in plain English. - [Machine Learning in AI: 6 Ways Computers Learn From Data](https://aiodic.com/machine-learning-in-ai/): Machine Learning on the Periodic Table of AI: six ways computers learn from data, from supervised and unsupervised learning to deep learning. - [AI Training Data: How Models Get Fed, Tuned and Tested](https://aiodic.com/ai-training-data/): Data & Training on the Periodic Table of AI: how datasets, pretraining, fine-tuning, RLHF, synthetic data and benchmarks shape AI models. - [AI Prompting: 20 Ways to Get Better Answers From AI](https://aiodic.com/ai-prompting-techniques/): Prompting on the Periodic Table of AI: 20 techniques, from few-shot examples and chain of thought to templates and evals, in plain English. - [AI Model Families: Who Makes GPT, Claude, Gemini, Llama and More](https://aiodic.com/ai-model-families/): Model Families on the Periodic Table of AI: who makes GPT, Claude, Gemini, Llama, DeepSeek, Qwen and nine more, and which ones you can download. - [AI Tools and Protocols: How AI Connects to the Outside World](https://aiodic.com/ai-tools-and-protocols/): Tools & Protocols on the Periodic Table of AI: how APIs, MCP, function calling, web search, RAG and vector databases connect AI to the world. - [AI Agents and Automation: When AI Stops Talking and Starts Doing](https://aiodic.com/ai-agents-and-automation/): Agents & Automation on the Periodic Table of AI: 18 elements, from tool use and the ReAct loop to computer use, subagents and autonomy levels. - [AI Applications: What People Actually Use AI For](https://aiodic.com/ai-applications/): Applications on the Periodic Table of AI: how people use AI for code, writing, translation, summaries, research, tutoring and customer support. - [AI Creative Tools: Image, Video, Music and Voice Generators Explained](https://aiodic.com/ai-creative-tools/): Creative Tools on the Periodic Table of AI: 15 image, video, music and voice generators explained, plus what creators should know about copyright. - [AI Safety and Ethics: 7 Problems Every AI User Should Understand](https://aiodic.com/ai-safety-and-ethics/): Safety & Ethics on the Periodic Table of AI: alignment, hallucination, bias, guardrails, human oversight, privacy and interpretability explained. - [The Frontier of AI: World Models, Robots and AGI](https://aiodic.com/ai-frontier/): Frontier on the Periodic Table of AI: world models, embodied AI, continual learning, AGI and superintelligence, and where the experts disagree. ## Pages - [Home](https://aiodic.com/): Aiodic is the Periodic Table of AI: 118 building blocks, from tokens to agents, each explained in plain English. Click any element to explore it. ## Elements - [Zero-shot](https://aiodic.com/element/zero-shot/): You describe the task and rely on the model's general ability. Fast and often good enough for common tasks. - [XML Tags](https://aiodic.com/element/xml-tags/): Wrapping sections in tags like or helps the model separate context from commands and makes outputs easy to parse. - [Workflow Automation](https://aiodic.com/element/workflow-automation/): Platforms like n8n or Zapier trigger AI steps when events happen, such as summarizing every new email or form entry. - [Web Search](https://aiodic.com/element/web-search/): Search gives the model current information beyond its training cutoff, with sources you can check. - [Writing Assistance](https://aiodic.com/element/writing-assistance/): From outlines to line edits, AI speeds up writing while you keep control of voice and judgment. - [World Models](https://aiodic.com/element/world-models/): World models predict what happens next in a physical or simulated space, a step toward planning in the real world. - [Transformer](https://aiodic.com/element/transformer/): Introduced in 2017, the transformer uses attention to weigh how every word relates to every other word, which made large-scale language models practical. - [Tokens](https://aiodic.com/element/tokens/): A token is often a word or part of a word. Pricing, speed, and context limits are all measured in tokens, so knowing them helps you plan prompts and costs. - [Supervised Learning](https://aiodic.com/element/supervised-learning/): The model sees inputs paired with correct answers, such as photos tagged cat or dog, and learns to predict the label for new inputs. - [Unsupervised Learning](https://aiodic.com/element/unsupervised-learning/): With no answers provided, the model groups, clusters, or compresses data on its own. Useful for discovering segments and hidden patterns. - [Reinforcement Learning](https://aiodic.com/element/reinforcement-learning/): An agent takes actions, receives rewards or penalties, and improves its strategy over time. Used in games, robotics, and training LLMs to follow instructions. - [Transfer Learning](https://aiodic.com/element/transfer-learning/): A model pretrained on broad data can be adapted to a narrow task with far less data and compute than starting from scratch. - [Pretraining](https://aiodic.com/element/pretraining/): The model learns general language and knowledge by predicting the next token across enormous amounts of text. This is the most expensive stage. - [RLHF](https://aiodic.com/element/rlhf/): People rate model responses, and those preferences train the model to be more helpful and less harmful. A key step in making chat assistants. - [Synthetic Data](https://aiodic.com/element/synthetic-data/): Models can produce examples to train or test other models. It fills gaps cheaply but needs checking to avoid compounding errors. - [System Prompt](https://aiodic.com/element/system-prompt/): Set by a developer or user, the system prompt defines role, tone, rules, and context before the conversation begins. - [Role Prompting](https://aiodic.com/element/role-prompting/): Telling the model to act as an editor, analyst, or tutor focuses vocabulary, priorities, and depth. - [Temperature](https://aiodic.com/element/temperature/): Low temperature gives focused, repeatable answers; high temperature gives more varied, creative ones. - [Structured Output](https://aiodic.com/element/structured-output/): Specifying an exact schema makes AI output reliable for software, spreadsheets, and automation. - [Self-consistency](https://aiodic.com/element/self-consistency/): Running the same reasoning question multiple times and voting reduces one-off mistakes. - [Self-critique](https://aiodic.com/element/self-critique/): A second pass that checks for errors, gaps, or rule violations often catches problems the first draft missed. - [Prompt Templates](https://aiodic.com/element/prompt-templates/): Templates keep quality consistent across repeated tasks, from product descriptions to chapter outlines. - [Tool Use](https://aiodic.com/element/tool-use/): With tools, a model can search, calculate, read files, or send messages instead of relying only on what it memorized. - [ReAct Loop](https://aiodic.com/element/react-loop/): The agent thinks about what to do, takes an action, reads the result, and uses it to decide the next move. - [Subagents](https://aiodic.com/element/subagents/): A main agent hands a narrow task to a subagent with its own clean context, then uses the result. - [Skills](https://aiodic.com/element/skills/): Skills bundle know-how and scripts for a task type, so the agent follows proven steps instead of improvising. - [Triggers](https://aiodic.com/element/triggers/): A new file, a scheduled time, a webhook, or a message can kick off an AI workflow without anyone pressing go. - [Scheduled Tasks](https://aiodic.com/element/scheduled-tasks/): Daily briefs, weekly reports, and recurring checks can run automatically at set times. - [RAG](https://aiodic.com/element/rag/): The system first finds relevant passages from your documents, then gives them to the model so answers are based on your content. - [Vector Database](https://aiodic.com/element/vector-database/): It stores embeddings and quickly finds the most similar items, which is the backbone of most RAG systems. - [Translation](https://aiodic.com/element/translation/): Modern models translate with context and tone, and can adapt register for formal or casual audiences. - [Summarization](https://aiodic.com/element/summarization/): AI turns reports, transcripts, and books into summaries at whatever length and angle you need. - [Research](https://aiodic.com/element/research/): AI searches, reads, compares sources, and drafts sourced reports, though claims still need checking. - [Superintelligence](https://aiodic.com/element/superintelligence/): A theoretical future system that exceeds human performance in nearly every domain, central to long-term safety debates. - [Privacy](https://aiodic.com/element/privacy/): Knowing what data you share with AI, where it's stored, and how it's used protects you and your customers. - [Qwen](https://aiodic.com/element/qwen/): A broad range of open and commercial models covering text, code, and vision. - [Stable Diffusion](https://aiodic.com/element/stable-diffusion/): Runs locally or in the cloud, with a large community of custom models and tools. - [Runway](https://aiodic.com/element/runway/): Tools for generating, extending, and editing video clips from text and images. - [Sora](https://aiodic.com/element/sora/): Generates video from text and image prompts. - [Veo](https://aiodic.com/element/veo/): Generates video clips, including versions with native audio, from prompts. - [Suno](https://aiodic.com/element/suno/): Creates full songs with vocals and instruments from a text description. - [Udio](https://aiodic.com/element/udio/): Generates songs and instrumentals across many genres from prompts. - [Neural Network](https://aiodic.com/element/neural-network/): Inspired loosely by the brain, a neural network adjusts millions or billions of connection strengths during training until its outputs match the examples it learns from. - [Language Model](https://aiodic.com/element/language-model/): LLMs like Claude and GPT generate text one token at a time. Scale and training let them write, summarize, code, and reason across many subjects. - [Parameters](https://aiodic.com/element/parameters/): Parameters are the adjustable weights set during training. More parameters can mean more capability, but also more cost to train and run. - [Inference](https://aiodic.com/element/inference/): Training builds the model once; inference happens every time someone sends a prompt. Most AI costs for users are inference costs. - [Mixture of Experts](https://aiodic.com/element/mixture-of-experts/): Only a few experts activate for each token, so the model can be very large while staying relatively fast and affordable to run. - [Machine Learning](https://aiodic.com/element/machine-learning/): Rather than hand-coding every rule, you give the system data and let it find the patterns. Most modern AI is built on machine learning. - [Prompt](https://aiodic.com/element/prompt/): Clear prompts state the task, context, audience, and desired format. Small wording changes can produce very different results. - [Prompt Chaining](https://aiodic.com/element/prompt-chaining/): The output of one prompt feeds the next, such as outline, then draft, then edit. Each step stays simple and checkable. - [Meta-prompting](https://aiodic.com/element/meta-prompting/): Ask the model to draft, critique, or optimize a prompt for a goal. A fast way to level up your own prompting. - [Prefilling](https://aiodic.com/element/prefilling/): Providing the first words of the response steers format and tone, such as beginning with an open brace for JSON. - [Grounding](https://aiodic.com/element/grounding/): Asking the model to answer only from provided documents and cite them reduces made-up facts. - [Iterative Refinement](https://aiodic.com/element/iterative-refinement/): Treat the first answer as a draft. Specific feedback on what to change gets closer to the goal each round. - [Planning](https://aiodic.com/element/planning/): Good agents outline a plan, execute it, and revise when a step fails or new information appears. - [Memory](https://aiodic.com/element/memory/): Memory lets an assistant recall preferences, past work, and project context so you don't repeat yourself. - [Multi-agent Systems](https://aiodic.com/element/multi-agent-systems/): Different agents take different roles, such as researcher, writer, and reviewer, and pass work between them. - [Orchestration](https://aiodic.com/element/orchestration/): An orchestrator decides what runs when, routes results, and handles failures across a larger system. - [Handoffs](https://aiodic.com/element/handoffs/): Clean handoffs include the goal, what's done, and what's next, so nothing is lost at the transfer. - [Observability](https://aiodic.com/element/observability/): Logs, traces, and cost tracking show each step an agent took, which makes debugging and trust possible. - [MCP](https://aiodic.com/element/mcp/): MCP lets assistants plug into apps like Drive, Slack, or your own servers through one common format, like a universal adapter. - [Hallucination](https://aiodic.com/element/hallucination/): Models can invent facts, quotes, or sources. Grounding, citations, and verification reduce the risk. - [Guardrails](https://aiodic.com/element/guardrails/): Filters, policies, and validation steps keep AI within safe and appropriate boundaries. - [Human in the Loop](https://aiodic.com/element/human-in-the-loop/): Keeping a person in key decisions catches errors and keeps accountability where it belongs. - [Interpretability](https://aiodic.com/element/interpretability/): Researchers study internal features and circuits to explain why a model produces a given output. - [Llama](https://aiodic.com/element/llama/): Downloadable weights let developers run and customize Llama models on their own hardware. - [Mistral](https://aiodic.com/element/mistral/): A European lab offering both open-weight and commercial models. - [Phi](https://aiodic.com/element/phi/): Compact models aimed at strong performance on modest hardware. - [Nova](https://aiodic.com/element/nova/): Models offered through Amazon's cloud for text, image, and video tasks. - [Kimi](https://aiodic.com/element/kimi/): Chinese models noted for long context and agentic coding. - [OLMo](https://aiodic.com/element/olmo/): Released with training data and code, making them useful for research and transparency. - [Midjourney](https://aiodic.com/element/midjourney/): Popular for stylized, high-aesthetic images from text prompts. - [OpenAI Images](https://aiodic.com/element/openai-images/): Creates and edits images conversationally, with strong prompt following and text rendering. - [Ideogram](https://aiodic.com/element/ideogram/): Well suited to posters, logos, and designs that need legible lettering. - [Kling](https://aiodic.com/element/kling/): A Chinese text-to-video and image-to-video generator. - [HeyGen](https://aiodic.com/element/heygen/): Creates presenter-style videos with digital avatars and translated lip-sync. - [Embeddings](https://aiodic.com/element/embeddings/): Similar ideas get similar embeddings, which lets software search by meaning instead of exact words. They power semantic search and RAG systems. - [Context Window](https://aiodic.com/element/context-window/): Everything in the conversation, including instructions, files, and replies, must fit in the context window. Larger windows allow whole books or codebases in one request. - [Diffusion Model](https://aiodic.com/element/diffusion-model/): Diffusion models start from random static and gradually refine it into a picture that matches the prompt. Most AI image and video tools use this approach. - [Deep Learning](https://aiodic.com/element/deep-learning/): Depth lets networks learn complex features, from edges in images to meaning in sentences. Deep learning drove the modern AI boom. - [Datasets](https://aiodic.com/element/datasets/): Data quality shapes model quality. Datasets can be text, images, audio, code, or structured records, and they carry the biases of their sources. - [Fine-tuning](https://aiodic.com/element/fine-tuning/): A pretrained model is trained further on focused examples so it adopts a style, format, or domain skill. - [Few-shot](https://aiodic.com/element/few-shot/): Two to five examples of input and ideal output teach format and style far better than description alone. - [Chain of Thought](https://aiodic.com/element/chain-of-thought/): Asking the model to think through a problem before answering improves accuracy on math, logic, and multi-step tasks. - [Context Engineering](https://aiodic.com/element/context-engineering/): Beyond wording, results depend on which documents, examples, and history you include, and what you leave out. - [Extended Thinking](https://aiodic.com/element/extended-thinking/): Some models can think at length internally before answering, which helps with complex analysis and planning. - [Evals](https://aiodic.com/element/evals/): A small suite of inputs with expected qualities shows whether a prompt change actually improved results. - [Computer Use](https://aiodic.com/element/computer-use/): The model sees screenshots and clicks or types like a person, so it can work in apps that have no API. - [Function Calling](https://aiodic.com/element/function-calling/): The model outputs a structured call, your code runs it, and the result goes back to the model to continue. - [Code Generation](https://aiodic.com/element/code-generation/): AI drafts functions, fixes bugs, writes tests, and explains unfamiliar code. - [Data Analysis](https://aiodic.com/element/data-analysis/): AI can clean spreadsheets, run calculations, build charts, and explain what the numbers mean. - [Tutoring](https://aiodic.com/element/tutoring/): AI explains concepts at your level, quizzes you, and adjusts when you get stuck. - [Customer Support](https://aiodic.com/element/customer-support/): AI handles common questions instantly and routes complex issues to people. - [Chatbots](https://aiodic.com/element/chatbots/): A chatbot wraps a model in a conversation, often with a specific role, knowledge base, and rules. - [Embodied AI](https://aiodic.com/element/embodied-ai/): Embodied systems combine perception, reasoning, and movement to act in the physical world. - [Continual Learning](https://aiodic.com/element/continual-learning/): Today's models mostly stay fixed after training. Continual learning would let them update safely from new experience. - [AGI](https://aiodic.com/element/agi/): A hypothetical AI able to match people across most intellectual tasks. Definitions and timelines are widely debated. - [GPT](https://aiodic.com/element/gpt/): The models behind ChatGPT, widely used for chat, coding, and multimodal tasks. - [Claude](https://aiodic.com/element/claude/): Known for long context, careful reasoning, writing, and coding, available in chat, apps, and the API. - [Gemini](https://aiodic.com/element/gemini/): Natively multimodal models integrated across Google products and available to developers. - [DeepSeek](https://aiodic.com/element/deepseek/): A Chinese lab known for efficient, openly released reasoning and chat models. - [Grok](https://aiodic.com/element/grok/): Models integrated with the X platform and offered through an API. - [Gemma](https://aiodic.com/element/gemma/): Lightweight models built from Gemini research for local and custom use. - [Command](https://aiodic.com/element/command/): Enterprise-focused models with strengths in retrieval and business workflows. - [GLM](https://aiodic.com/element/glm/): Chinese open and commercial models for chat, code, and agents. - [Flux](https://aiodic.com/element/flux/): Image models known for photorealism and prompt accuracy, in open and commercial versions. - [Adobe Firefly](https://aiodic.com/element/adobe-firefly/): Built into Photoshop and other Adobe apps for generating and editing images and more. - [ElevenLabs](https://aiodic.com/element/elevenlabs/): Realistic text-to-speech, voice cloning, and dubbing, popular for audiobooks. - [Canva AI](https://aiodic.com/element/canva-ai/): Generates images, text, and layouts for social posts, covers, and presentations. - [Artificial Intelligence](https://aiodic.com/element/artificial-intelligence/): The umbrella term for systems that learn, reason, perceive, or create. Everything else on this table is a part, method, product, or concern of AI. - [Benchmarks](https://aiodic.com/element/benchmarks/): Benchmarks score models on math, coding, reasoning, and more. They are useful signals but can be gamed or miss real-world performance. - [Agent](https://aiodic.com/element/agent/): Agents plan steps, use tools, check results, and keep going until a task is done, rather than answering once and stopping. - [Browser Agents](https://aiodic.com/element/browser-agents/): Browser agents fill forms, gather information, and complete multi-page tasks inside a web browser. - [Coding Agents](https://aiodic.com/element/coding-agents/): Tools like Claude Code work in a real project, editing files, running tests, and iterating until the code works. - [Autonomy Levels](https://aiodic.com/element/autonomy-levels/): From suggest-only to fully independent, setting autonomy decides when a human must approve an action. - [API](https://aiodic.com/element/api/): Developers send prompts and receive responses through an API, which lets AI be built into apps, sites, and scripts. - [Alignment](https://aiodic.com/element/alignment/): Alignment research aims to ensure systems are helpful, honest, and harmless, even as they grow more capable. - [Bias](https://aiodic.com/element/bias/): Models can reflect stereotypes or gaps in their training data, which matters in hiring, lending, and more.