[{"type":"Article","title":"AI's Energy Appetite Drives Nvidia and Amazon to Pour Billions Into Power Infrastructure","description":"Amazon secures multi-gigawatt power agreements for AI data centers as nuclear and gas facilities come online.","url":"/article/2026-08-09-ais-energy-appetite-drives-nvidia-and-amazon-to-pour-billion/"},{"type":"Article","title":"Anthropic Is Turning Claude Code's Auto Mode on by Default","description":"The change makes the agentic coding feature opt-out rather than opt-in for developers using the tool.","url":"/article/2026-08-09-anthropic-is-turning-claude-codes-auto-mode-on-by-default/"},{"type":"Article","title":"Google's DiffusionGemma Shows Text Diffusion Models Can Skip Training From Scratch","description":"The approach demonstrates that fine-tuning existing pre-trained models can yield functional text diffusion systems.","url":"/article/2026-08-09-googles-diffusiongemma-proves-you-dont-need-to-train-from-sc/"},{"type":"Article","title":"DeepMind's WeatherNext AI Model Gives Hurricane Forecasters Extra Day of Lead Time","description":"The model predicted Hurricane Melissa would hit Jamaica as a Category 5 with 80 percent confidence five days before landfall.","url":"/article/2026-08-09-deepminds-hurricane-breakthrough-has-surprised-weather-scientists/"},{"type":"Article","title":"DeepMind's WeatherNext AI Model Gives Hurricane Forecasters Extra Day of Lead Time","description":"The model predicted Hurricane Melissa would hit Jamaica as a Category 5 with 80 percent confidence five days before landfall.","url":"/article/2026-08-09-deepminds-weathernext-model-gives-hurricane-forecasters-extra-day/"},{"type":"Guide · Expert","title":"Key Papers on the Road to AGI","description":"A curated reading list tracing the arc from the transformer to scaling laws, RLHF, and reasoning models — plus how to read an ML paper efficiently.","url":"/advanced/key-papers/","keywords":"AI research papers, transformer, scaling laws, RLHF, chain of thought, reasoning models, how to read ML papers, AGI reading list"},{"type":"Guide · Intermediate","title":"Prompting That Works","description":"What actually improves AI prompts: context, examples, structure, and iteration — the myths worth dropping, and when prompting stops being the fix.","url":"/guides/prompting-that-works/","keywords":"prompt engineering, prompting techniques, few-shot prompting, system prompt, AI context, prompt iteration, LLM prompting"},{"type":"Guide · Beginner","title":"What Is AGI, Actually?","description":"AI, AGI, and superintelligence in plain English: what today's systems do well, what they fail at, and why no one agrees on what AGI means.","url":"/start/what-is-agi/","keywords":"what is AGI, artificial general intelligence, AI vs AGI, superintelligence, AGI definition, AGI explained, AGI meaning"},{"type":"Guide · Expert","title":"The Benchmark Landscape","description":"How AI evaluation evolved from static QA to contaminated leaderboards to private, agentic tests — and what a healthy benchmark diet looks like in 2026.","url":"/advanced/benchmark-landscape/","keywords":"AI benchmarks, evaluation, contamination, SWE-bench, GAIA, ARC-AGI, FrontierMath, Humanity's Last Exam, agentic evals, private evals"},{"type":"Guide · Intermediate","title":"Context Windows and Tokens, Explained","description":"Tokens, context windows, and why AI models forget: what the limit is, what happens when you hit it, the long-context tradeoffs, and practical habits.","url":"/guides/context-windows-and-tokens/","keywords":"context window, tokens, tokenizer, long context, LLM memory, lost in the middle, context length, token limit"},{"type":"Guide · Beginner","title":"Choosing Your First AI Chatbot","description":"A plain-English guide to picking your first AI chatbot: Claude, ChatGPT, Gemini, Copilot, and Perplexity, how to choose, and how to stay private.","url":"/start/your-first-ai-chatbot/","keywords":"best AI chatbot, first AI chatbot, Claude, ChatGPT, Gemini, Copilot, Perplexity, free AI chatbot, how to choose AI"},{"type":"Guide · Beginner","title":"Free vs Paid AI: What You Actually Get","description":"What free AI chatbots include, what a paid plan adds, and when the upgrade is worth it. For many people, the free tier is genuinely enough.","url":"/start/free-vs-paid-ai/","keywords":"free vs paid AI, ChatGPT free, Claude free, AI subscription cost, is paid AI worth it, AI pricing, free AI tier"},{"type":"Guide · Intermediate","title":"Local Models: Hardware and Quantization","description":"What determines which local AI models you can run: memory first, unified memory vs VRAM, what Q4 and Q8 quantization mean, GGUF, and realistic tiers.","url":"/guides/local-models-hardware/","keywords":"local LLM, quantization, GGUF, Q4, Q8, unified memory, VRAM, llama.cpp, Ollama, MLX, run models locally"},{"type":"Guide · Expert","title":"Scaling Laws and the Road to AGI","description":"Kaplan and Chinchilla explained precisely, what scaling laws predict and don't, inference-time compute, and the 'wall' debate as of August 2026.","url":"/advanced/scaling-laws/","keywords":"scaling laws, Chinchilla, Kaplan, compute-optimal, inference-time compute, reasoning models, data wall, AI scaling debate, AGI"},{"type":"Guide · Expert","title":"A Map of Alignment Research","description":"A map of AI alignment research: outer and inner alignment, RLHF and Constitutional AI, interpretability, control, evals, and who works on what.","url":"/advanced/alignment-research-map/","keywords":"AI alignment, RLHF, Constitutional AI, interpretability, sparse autoencoders, AI control, scalable oversight, deceptive alignment, red-teaming, evals"},{"type":"Guide · Intermediate","title":"RAG, Explained","description":"How RAG works in plain terms: embeddings, retrieval, and chunking; where it shines and disappoints; RAG vs long context; and when to build vs use built-in.","url":"/guides/rag-explained/","keywords":"RAG, retrieval augmented generation, embeddings, vector database, chunking, semantic search, RAG vs long context, grounding"},{"type":"Guide · Beginner","title":"Running AI on Your Own Computer, Explained","description":"Why people run AI models on their own computer, what open weights means, the hardware you need, and free tools like LM Studio and Ollama.","url":"/start/run-ai-locally/","keywords":"run AI locally, local LLM, open weights, LM Studio, Ollama, offline AI, private AI, run AI on your computer"},{"type":"Guide · Intermediate","title":"Agents and Tool Use, Explained","description":"From chatbot to agent: how tool use and function calling work, the reason-act-observe loop, coding agents, MCP, and what agents still get wrong.","url":"/guides/agents-and-tool-use/","keywords":"AI agents, tool use, function calling, agent loop, coding agents, MCP, model context protocol, autonomous agents"},{"type":"Guide · Beginner","title":"AI Safety Basics: Scams, Hallucinations, and Good Habits","description":"How to spot AI hallucinations, defend your family from voice-cloning scams, protect your privacy, guide kids, and what AI safety really means.","url":"/start/ai-safety-basics/","keywords":"AI safety, AI hallucinations, voice cloning scam, deepfake scam, AI privacy, kids and AI, AI alignment, AI scams"},{"type":"Guide · Expert","title":"Inference Optimization: Quantization, Speculative Decoding, Batching","description":"Why inference economics rule everything: the KV cache and memory wall, quantization, speculative decoding, continuous batching, MoE, and the serving stack.","url":"/advanced/inference-optimization/","keywords":"LLM inference, quantization, speculative decoding, KV cache, continuous batching, PagedAttention, vLLM, SGLang, TensorRT-LLM, mixture of experts"},{"type":"Guide · Expert","title":"Agent Protocols and Interoperability (MCP and Friends)","description":"Why agent protocols emerged, MCP in depth, agent-to-agent interoperability, and the prompt-injection and tool-poisoning security surface.","url":"/advanced/agent-protocols/","keywords":"Model Context Protocol, MCP, agent protocols, A2A, agent interoperability, prompt injection, tool poisoning, AI agents, tool use, JSON-RPC"},{"type":"Guide · Intermediate","title":"How to Read AI Benchmarks Without Being Fooled","description":"How to read AI benchmarks without being fooled: saturation, contamination, why beats-X-on-Y headlines mislead, and a checklist for model announcements.","url":"/guides/reading-benchmarks/","keywords":"AI benchmarks, benchmark contamination, benchmark saturation, MMLU, SWE-bench, model evaluation, independent evals, model announcements"},{"type":"Guide · Intermediate","title":"Fine-Tuning vs RAG vs Prompting","description":"Fine-tuning vs RAG vs prompting: what each changes, costs, and fails at; the wrong reasons to fine-tune; and system prompts plus few-shot as the middle.","url":"/guides/fine-tuning-vs-rag-vs-prompting/","keywords":"fine-tuning, RAG, prompting, LoRA, QLoRA, system prompt, few-shot, when to fine-tune, decision framework"},{"type":"Model","title":"Anthropic Claude Opus 5","description":"Anthropic's Opus-tier model for complex agentic coding and enterprise work, positioned near Fable 5's intelligence at roughly half the price. Ships with a 1M-token context window.","url":"/models/#claude-opus-5","keywords":"Multimodal closed"},{"type":"Model","title":"Anthropic Claude Sonnet 5","description":"The balanced tier in the Claude 5 family, tuned for the best combination of speed and intelligence across coding and agent workloads. Carries a 1M-token context window.","url":"/models/#claude-sonnet-5","keywords":"Multimodal closed"},{"type":"Model","title":"Anthropic Claude Fable 5","description":"Anthropic's most capable widely released model and the first generally available Mythos-class tier, aimed at long-running agents. Priced at $10/$50 per million input/output tokens.","url":"/models/#claude-fable-5","keywords":"Multimodal closed"},{"type":"Model","title":"Anthropic Claude Opus 4.8","description":"The prior Opus flagship, now a legacy model succeeded by Opus 5 but still available via the Claude API. Introduced user-facing effort control and a faster inference mode.","url":"/models/#claude-opus-4-8","keywords":"Multimodal closed"},{"type":"Model","title":"Anthropic Claude Haiku 4.5","description":"Anthropic's fastest model, offering near-frontier intelligence at low cost with a 200K-token context window. Supports optional extended thinking.","url":"/models/#claude-haiku-4-5","keywords":"Multimodal closed"},{"type":"Model","title":"OpenAI GPT-5.6","description":"OpenAI's current flagship family, shipping in three tiers - Luna, Terra, and Sol - from most cost-efficient to most capable. Sol is positioned as OpenAI's strongest coding and vision model to date.","url":"/models/#gpt-5-6","keywords":"Multimodal closed"},{"type":"Model","title":"OpenAI gpt-oss-120b","description":"OpenAI's larger open-weight model (about 117B parameters), released under Apache 2.0 for local and self-hosted use with a focus on reasoning tasks.","url":"/models/#gpt-oss-120b","keywords":"Apache 2.0 Text open weights"},{"type":"Model","title":"OpenAI gpt-oss-20b","description":"OpenAI's smaller open-weight model (about 22B parameters) under Apache 2.0, designed to run efficiently on modest hardware.","url":"/models/#gpt-oss-20b","keywords":"Apache 2.0 Text open weights"},{"type":"Model","title":"Google DeepMind Gemini 3.6 Flash","description":"Google's high-throughput workhorse model, tuned for lower latency and roughly 17% fewer output tokens than 3.5 Flash. Generally available for production use.","url":"/models/#gemini-3-6-flash","keywords":"Multimodal closed"},{"type":"Model","title":"Google DeepMind Gemini 3.5 Flash-Lite","description":"The fastest, lowest-cost model in Google's Gemini 3.5 line, aimed at high-volume, latency-sensitive workloads.","url":"/models/#gemini-3-5-flash-lite","keywords":"Multimodal closed"},{"type":"Model","title":"Google DeepMind Gemma 4","description":"Google DeepMind's open-weight family (E2B, E4B, 26B MoE, and 31B dense) built from the same research as Gemini 3, now shipped under Apache 2.0. Handles text, images, audio, and video with up to 256K context.","url":"/models/#gemma-4","keywords":"Apache 2.0 Multimodal open weights"},{"type":"Model","title":"Meta AI Llama 4 Maverick","description":"Meta's larger open-weight Llama 4 model, a mixture-of-experts design (about 400B total, 17B active parameters) with native text-and-image input.","url":"/models/#llama-4-maverick","keywords":"Llama 4 Community License Multimodal open weights"},{"type":"Model","title":"Meta AI Llama 4 Scout","description":"The smaller Llama 4 model that fits on a single high-end GPU, with 17B active parameters and a very long context window.","url":"/models/#llama-4-scout","keywords":"Llama 4 Community License Multimodal open weights"},{"type":"Model","title":"DeepSeek DeepSeek-V4-Pro","description":"DeepSeek's flagship open-weight model, a 1.6T-parameter mixture-of-experts release under the MIT license with image input and a 1M-token context via API.","url":"/models/#deepseek-v4-pro","keywords":"MIT Multimodal open weights"},{"type":"Model","title":"DeepSeek DeepSeek-V4-Flash","description":"The smaller, faster text-only member of the DeepSeek V4 family (about 291B parameters), MIT-licensed and tuned for agentic and coding workloads.","url":"/models/#deepseek-v4-flash","keywords":"MIT Text open weights"},{"type":"Model","title":"Alibaba (Qwen) Qwen3.8-Max","description":"Alibaba's flagship Qwen model, a 2.4T-parameter mixture-of-experts system (about 95B active) with a 1M-token context and native text-plus-vision input. Currently API-only; Alibaba has announced an open-weights release.","url":"/models/#qwen3-8-max","keywords":"Multimodal closed"},{"type":"Model","title":"Alibaba (Qwen) Qwen3.6-27B","description":"An open-weight (Apache 2.0) Qwen model aimed at coding, with a vision encoder for multimodal input. Sized at 27B parameters to run on a single high-end GPU.","url":"/models/#qwen3-6-27b","keywords":"Apache 2.0 Multimodal open weights"},{"type":"Model","title":"Mistral AI Mistral Large 3","description":"Mistral's open-weight flagship, a 675B-parameter sparse mixture-of-experts model (41B active) under Apache 2.0 with an added vision encoder for image understanding.","url":"/models/#mistral-large-3","keywords":"Apache 2.0 Multimodal open weights"},{"type":"Model","title":"xAI Grok 4.6","description":"xAI's 1.5T-parameter frontier model, a refinement of Grok 4.5 through improved fine-tuning and reinforcement learning rather than a scale increase. Closed-weight, available through the xAI API and Grok apps.","url":"/models/#grok-4-6","keywords":"Multimodal closed"},{"type":"Model","title":"MiniMax MiniMax M2.7","description":"An open-weight 230B-parameter mixture-of-experts model (about 10B active) built for agentic and coding workflows, positioned as a low-cost frontier option.","url":"/models/#minimax-m2-7","keywords":"Text open weights"},{"type":"Model","title":"Moonshot AI Kimi K3","description":"Moonshot AI's open-weight mixture-of-experts model with 2.8T total parameters (104B active) and a 1M-token context, among the largest open models released. Multimodal via a native vision encoder.","url":"/models/#kimi-k3","keywords":"Kimi K3 License Multimodal open weights"},{"type":"Lab","title":"Anthropic","description":"AI safety and research company that develops the Claude family of frontier models, with an emphasis on interpretability, alignment, and reliable agentic systems.","url":"/labs/#anthropic"},{"type":"Lab","title":"OpenAI","description":"Research company behind the GPT models and ChatGPT, building general-purpose AI systems along with open-weight releases such as the gpt-oss models.","url":"/labs/#openai"},{"type":"Lab","title":"Google DeepMind","description":"Google's AI research lab, formed in 2023 by merging DeepMind (founded 2010) and Google Brain. Develops the Gemini models and the open-weight Gemma family.","url":"/labs/#google-deepmind"},{"type":"Lab","title":"Meta AI","description":"Meta's AI division, tracing to FAIR (founded 2013) and now organized under Meta Superintelligence Labs. Develops the open-weight Llama family.","url":"/labs/#meta-ai"},{"type":"Lab","title":"xAI","description":"Elon Musk's AI company, developer of the Grok models, integrated with the X platform and trained on its Colossus supercomputing clusters.","url":"/labs/#xai"},{"type":"Lab","title":"Mistral AI","description":"Paris-based lab known for efficient open-weight models, including the Mistral and Ministral families and the Le Chat assistant.","url":"/labs/#mistral-ai"},{"type":"Lab","title":"DeepSeek","description":"Hangzhou-based lab backed by the High-Flyer hedge fund, known for MIT-licensed open-weight mixture-of-experts models such as the DeepSeek V4 family.","url":"/labs/#deepseek"},{"type":"Lab","title":"Alibaba (Qwen)","description":"Alibaba's large-model team behind the Qwen (Tongyi Qianwen) family, first released in 2023, spanning frontier Max-class models and open-weight releases.","url":"/labs/#alibaba-qwen"},{"type":"Lab","title":"Moonshot AI","description":"Beijing-based startup that builds the Kimi chatbot and the Kimi series of large models, including large open-weight mixture-of-experts releases.","url":"/labs/#moonshot-ai"},{"type":"Lab","title":"MiniMax","description":"Shanghai-based AI company, publicly listed in Hong Kong in 2026, developing the MiniMax model series for agentic, coding, and multimodal tasks.","url":"/labs/#minimax"},{"type":"Lab","title":"Cohere","description":"Toronto-based enterprise AI company building the Command model family and retrieval tooling aimed at regulated industries and the public sector.","url":"/labs/#cohere"},{"type":"Lab","title":"Allen Institute for AI (Ai2)","description":"Seattle-based nonprofit founded by Paul Allen, known for fully open models such as the OLMo family, publishing weights, data, and training code.","url":"/labs/#ai2"},{"type":"Tool","title":"Claude","description":"General-purpose AI assistant from Anthropic for conversation, writing, coding, and document analysis.","url":"/tools/#claude","keywords":"Chat apps"},{"type":"Tool","title":"ChatGPT","description":"OpenAI's conversational assistant with text, image, voice, and web-browsing features.","url":"/tools/#chatgpt","keywords":"Chat apps"},{"type":"Tool","title":"Gemini","description":"Google's AI assistant, integrated with Google Search and Workspace apps.","url":"/tools/#gemini","keywords":"Chat apps"},{"type":"Tool","title":"Microsoft Copilot","description":"Microsoft's AI assistant for chat, web search, and image generation, integrated across Windows and Microsoft 365.","url":"/tools/#microsoft-copilot","keywords":"Chat apps"},{"type":"Tool","title":"Perplexity","description":"AI answer engine that responds to questions with cited web sources.","url":"/tools/#perplexity","keywords":"Chat apps"},{"type":"Tool","title":"Grok","description":"Conversational AI from xAI with real-time access to web and X content.","url":"/tools/#grok","keywords":"Chat apps"},{"type":"Tool","title":"DeepSeek","description":"Chat assistant from DeepSeek AI that runs the company's open-weight models.","url":"/tools/#deepseek","keywords":"Chat apps"},{"type":"Tool","title":"LM Studio","description":"Desktop application for discovering, downloading, and running local LLMs with a chat UI and an OpenAI-compatible local server.","url":"/tools/#lm-studio","keywords":"Local runners"},{"type":"Tool","title":"Ollama","description":"Tool for downloading and running open-weight models locally, with a command line and a local API.","url":"/tools/#ollama","keywords":"Local runners"},{"type":"Tool","title":"llama.cpp","description":"C/C++ library and command-line tools for running LLM inference locally across a wide range of hardware.","url":"/tools/#llama-cpp","keywords":"Local runners"},{"type":"Tool","title":"Jan","description":"Open-source desktop app to run LLMs offline, with local chat history and an OpenAI-compatible server.","url":"/tools/#jan","keywords":"Local runners"},{"type":"Tool","title":"GPT4All","description":"Desktop app from Nomic AI for running open-source LLMs privately on your own device, including local document chat.","url":"/tools/#gpt4all","keywords":"Local runners"},{"type":"Tool","title":"Claude Code","description":"Anthropic's agentic coding tool that works in the terminal, IDEs, and the web to build, debug, and ship code across a codebase.","url":"/tools/#claude-code","keywords":"Coding agents & dev tools"},{"type":"Tool","title":"Cursor","description":"AI-native code editor, built as a VS Code fork, with in-editor chat, autocomplete, and agentic multi-file editing.","url":"/tools/#cursor","keywords":"Coding agents & dev tools"},{"type":"Tool","title":"GitHub Copilot","description":"AI pair-programmer from GitHub offering code completion and chat across major IDEs and the command line.","url":"/tools/#github-copilot","keywords":"Coding agents & dev tools"},{"type":"Tool","title":"OpenAI Codex CLI","description":"OpenAI's open-source terminal coding agent that runs tasks in a local sandbox.","url":"/tools/#openai-codex-cli","keywords":"Coding agents & dev tools"},{"type":"Tool","title":"Aider","description":"Open-source command-line coding assistant that pairs with your local git repository; you bring your own model API key.","url":"/tools/#aider","keywords":"Coding agents & dev tools"},{"type":"Tool","title":"Cline","description":"Open-source autonomous coding agent that runs as a sidebar in VS Code and other editors; you bring your own model API key.","url":"/tools/#cline","keywords":"Coding agents & dev tools"},{"type":"Tool","title":"Zed","description":"Open-source, high-performance code editor written in Rust with built-in AI assistance and collaboration features.","url":"/tools/#zed","keywords":"Coding agents & dev tools"},{"type":"Tool","title":"Midjourney","description":"Subscription image-generation service, accessible through a web app and Discord.","url":"/tools/#midjourney","keywords":"Creative tools"},{"type":"Tool","title":"Google Veo","description":"Google DeepMind's text-to-video and image-to-video model with native audio generation, available through Gemini, Flow, and the Gemini API.","url":"/tools/#google-veo","keywords":"Creative tools"},{"type":"Tool","title":"Google Imagen","description":"Google DeepMind's text-to-image model, available through Gemini, Google AI Studio, and the Gemini API.","url":"/tools/#google-imagen","keywords":"Creative tools"},{"type":"Tool","title":"ElevenLabs","description":"AI audio platform for text-to-speech, voice cloning, dubbing, and music generation.","url":"/tools/#elevenlabs","keywords":"Creative tools"},{"type":"Tool","title":"Suno","description":"AI music generator that creates songs with vocals and instrumentation from text prompts.","url":"/tools/#suno","keywords":"Creative tools"},{"type":"Tool","title":"Runway","description":"AI platform for video generation and editing, including text-to-video and image-to-video tools.","url":"/tools/#runway","keywords":"Creative tools"},{"type":"Tool","title":"Ideogram","description":"AI image generator noted for rendering legible text within images.","url":"/tools/#ideogram","keywords":"Creative tools"},{"type":"Tool","title":"Claude Agent SDK","description":"Anthropic's SDK for building agents on the same tools, agent loop, and context management that power Claude Code; formerly the Claude Code SDK.","url":"/tools/#claude-agent-sdk","keywords":"Agent frameworks & libraries"},{"type":"Tool","title":"LangChain","description":"Open-source framework for building applications and agents with LLMs, with integrations across many model and tool providers.","url":"/tools/#langchain","keywords":"Agent frameworks & libraries"},{"type":"Tool","title":"LangGraph","description":"Open-source library from the LangChain team for building stateful, multi-step agent workflows modeled as graphs.","url":"/tools/#langgraph","keywords":"Agent frameworks & libraries"},{"type":"Tool","title":"CrewAI","description":"Open-source Python framework for orchestrating role-based, collaborating AI agents.","url":"/tools/#crewai","keywords":"Agent frameworks & libraries"},{"type":"Tool","title":"OpenAI Agents SDK","description":"OpenAI's lightweight open-source SDK for building multi-agent workflows with handoffs, guardrails, and tracing.","url":"/tools/#openai-agents-sdk","keywords":"Agent frameworks & libraries"},{"type":"Tool","title":"LlamaIndex","description":"Open-source data framework for connecting LLMs and agents to external data sources for retrieval and workflows.","url":"/tools/#llamaindex","keywords":"Agent frameworks & libraries"},{"type":"Tool","title":"Google Agent Development Kit (ADK)","description":"Google's open-source Agent Development Kit for building, testing, and deploying agents, with support for multiple languages.","url":"/tools/#google-adk","keywords":"Agent frameworks & libraries"},{"type":"Glossary","title":"Agent","description":"A software system that uses an AI model to take actions on its own toward a goal, such as browsing the web, running code, or using tools, rather than just answering a single question. Agents typically work in a loop: decide, act, observe the result, and try again.","url":"/glossary/#agent"},{"type":"Glossary","title":"AGI (Artificial General Intelligence)","description":"A hypothetical AI that could understand and perform any intellectual task a human can, across many domains rather than one narrow specialty. There is no agreed-upon test for when AGI would be reached, and experts disagree on whether or when it will happen.","url":"/glossary/#agi-artificial-general-intelligence"},{"type":"Glossary","title":"Alignment","description":"The effort to make AI systems pursue the goals and values their designers and users actually intend, and to behave safely and honestly. Misalignment is when a system optimizes for something subtly different from what people want.","url":"/glossary/#alignment"},{"type":"Glossary","title":"ASI (Artificial Superintelligence)","description":"A hypothetical AI that would be far more capable than the best humans at essentially everything. It is a step beyond AGI and remains speculative.","url":"/glossary/#asi-artificial-superintelligence"},{"type":"Glossary","title":"Attention","description":"A mechanism inside modern AI models that lets them weigh which earlier words or pieces of input matter most when producing the next piece of output. It is the core idea behind the transformer, the architecture used by most large language models.","url":"/glossary/#attention"},{"type":"Glossary","title":"Backpropagation","description":"The core training algorithm for neural networks. It works out how much each internal setting contributed to the model's errors and nudges all of them in the direction that reduces those errors, repeated over huge amounts of data.","url":"/glossary/#backpropagation"},{"type":"Glossary","title":"Benchmark","description":"A standardized test used to measure and compare how well AI models perform on a specific kind of task, such as math, coding, or general knowledge. Benchmarks help track progress, but a high score does not always translate to real-world usefulness.","url":"/glossary/#benchmark"},{"type":"Glossary","title":"Chain of thought","description":"A technique where a model works through a problem step by step in writing before giving its final answer, much like showing your work in math. It often improves accuracy on reasoning-heavy tasks.","url":"/glossary/#chain-of-thought"},{"type":"Glossary","title":"Chatbot","description":"A program you interact with by typing or speaking in everyday language, and that responds conversationally. Modern chatbots like ChatGPT and Claude are powered by large language models.","url":"/glossary/#chatbot"},{"type":"Glossary","title":"Compute","description":"The raw computing power, usually measured as processor time on chips like GPUs, needed to train or run an AI model. More compute generally allows bigger models and more training, and it is one of the main costs and bottlenecks in AI.","url":"/glossary/#compute"},{"type":"Glossary","title":"Context window","description":"The amount of text, measured in tokens, that a model can take in and consider at once, including both your input and its own response. Anything beyond that limit is dropped, so a larger context window lets a model work with longer documents or conversations.","url":"/glossary/#context-window"},{"type":"Glossary","title":"Deep learning","description":"A type of machine learning that uses neural networks with many layers to learn patterns directly from large amounts of data. It powers most modern AI, including image recognition and large language models.","url":"/glossary/#deep-learning"},{"type":"Glossary","title":"Diffusion model","description":"A type of AI model that generates images or other data by starting with random noise and gradually refining it into a coherent result. It is the technology behind many popular image generators.","url":"/glossary/#diffusion-model"},{"type":"Glossary","title":"Distillation","description":"A technique for training a smaller, cheaper student model to imitate a larger, more capable teacher model. The goal is to keep much of the quality while cutting the cost and speed of running it.","url":"/glossary/#distillation"},{"type":"Glossary","title":"Embedding","description":"A way of representing words, images, or other data as a list of numbers so that items with similar meaning end up close together. Embeddings let software measure similarity and are a building block of search and recommendation systems.","url":"/glossary/#embedding"},{"type":"Glossary","title":"Emergent abilities","description":"Skills that appear in large models but were not present in smaller ones, seeming to switch on once a model reaches a certain size or amount of training. Researchers debate how real and how predictable these jumps actually are.","url":"/glossary/#emergent-abilities"},{"type":"Glossary","title":"Few-shot learning","description":"Giving a model a handful of examples of a task inside the prompt so it can follow the pattern, without any additional training. It contrasts with zero-shot, where no examples are provided.","url":"/glossary/#few-shot-learning"},{"type":"Glossary","title":"Fine-tuning","description":"Taking an already-trained model and training it further on a narrower set of examples so it does better at a specific task or adopts a particular style. It is usually far cheaper than training a model from scratch.","url":"/glossary/#fine-tuning"},{"type":"Glossary","title":"Foundation model","description":"A large model trained on broad data that can be adapted to many different tasks, serving as a base that others build on. Large language models are the best-known example.","url":"/glossary/#foundation-model"},{"type":"Glossary","title":"Frontier model","description":"One of the most capable AI models available at a given time, typically from a leading lab and trained at very large scale. The term comes up often in discussions of AI safety and regulation.","url":"/glossary/#frontier-model"},{"type":"Glossary","title":"Generative AI","description":"AI that creates new content, such as text, images, audio, video, or code, rather than just classifying or scoring existing data. Chatbots and image generators are common examples.","url":"/glossary/#generative-ai"},{"type":"Glossary","title":"GPU","description":"Short for graphics processing unit, a type of chip originally built for rendering graphics that turns out to be very good at the parallel math AI models need. GPUs are the main hardware used to train and run modern AI.","url":"/glossary/#gpu"},{"type":"Glossary","title":"Guardrails","description":"Rules, filters, or added safety layers that keep an AI system from producing harmful or off-limits outputs. They sit around the model rather than being part of how it fundamentally works.","url":"/glossary/#guardrails"},{"type":"Glossary","title":"Hallucination","description":"When an AI model states something false or made-up as if it were true, often fluently and confidently. It happens because models predict plausible-sounding text rather than looking up verified facts.","url":"/glossary/#hallucination"},{"type":"Glossary","title":"Inference","description":"The act of running a trained model to get an answer, as opposed to training it. Every time you send a prompt and get a response, that is inference.","url":"/glossary/#inference"},{"type":"Glossary","title":"Jailbreak","description":"A prompt or trick designed to get an AI model to bypass its safety rules and produce content it is meant to refuse. Labs continually patch known jailbreaks as they are discovered.","url":"/glossary/#jailbreak"},{"type":"Glossary","title":"Knowledge cutoff","description":"The date after which a model has no built-in knowledge, because its training data only goes up to that point. Events after the cutoff are unknown to the model unless it is given that information or can look it up.","url":"/glossary/#knowledge-cutoff"},{"type":"Glossary","title":"Large language model (LLM)","description":"An AI model trained on huge amounts of text to predict and generate language, which lets it answer questions, write, summarize, and more. Often abbreviated LLM, these models power today's leading chatbots.","url":"/glossary/#large-language-model-llm"},{"type":"Glossary","title":"Latency","description":"The delay between sending a request to an AI model and getting a response back. Lower latency means the system feels faster and more responsive.","url":"/glossary/#latency"},{"type":"Glossary","title":"Machine learning","description":"A branch of AI where systems learn patterns from data and improve with experience, instead of being explicitly programmed with rules for every case. Deep learning is one powerful kind of machine learning.","url":"/glossary/#machine-learning"},{"type":"Glossary","title":"MCP (Model Context Protocol)","description":"An open standard, introduced by Anthropic in late 2024, for connecting AI models to outside tools and data sources in a consistent way. It lets developers plug an assistant into things like files, databases, and apps without custom wiring for each one.","url":"/glossary/#mcp-model-context-protocol"},{"type":"Glossary","title":"Mixture of experts","description":"A model design that splits the network into many specialized sub-models, or experts, and for each input uses only a few of them. This keeps a model large in total knowledge while making each response cheaper to compute.","url":"/glossary/#mixture-of-experts"},{"type":"Glossary","title":"Multimodal","description":"Describes an AI model that can handle more than one type of input or output, such as text, images, audio, and video, rather than text alone. Most leading models today are multimodal.","url":"/glossary/#multimodal"},{"type":"Glossary","title":"Neural network","description":"A computing system loosely inspired by the brain, made of layers of simple connected units whose numeric settings are tuned during training. It is the basic structure underneath deep learning and modern AI.","url":"/glossary/#neural-network"},{"type":"Glossary","title":"Open weights","description":"When a model's trained parameters are released publicly so anyone can download, run, and adapt it. This differs from fully open source, since the training data and code may not be shared, and from closed models offered only through an online service.","url":"/glossary/#open-weights"},{"type":"Glossary","title":"Overfitting","description":"When a model learns its training examples too closely, including their quirks and noise, and as a result performs worse on new, unseen data. Avoiding it is a central concern in machine learning.","url":"/glossary/#overfitting"},{"type":"Glossary","title":"Parameter","description":"One of the internal numeric values a model adjusts during training; together they store what the model has learned. Modern large models have billions or more, and the count is a rough, imperfect indicator of capacity.","url":"/glossary/#parameter"},{"type":"Glossary","title":"Pre-training","description":"The first and largest training stage, where a model learns general patterns from a huge, broad dataset. For language models this usually means learning to predict the next token across vast amounts of text, with later stages like fine-tuning specializing it.","url":"/glossary/#pre-training"},{"type":"Glossary","title":"Prompt","description":"The input or instruction you give an AI model to get a response, such as a question, a request, or any text you type in. The wording of a prompt can strongly affect the quality of the answer.","url":"/glossary/#prompt"},{"type":"Glossary","title":"Prompt engineering","description":"The practice of carefully wording and structuring prompts to get better, more reliable results from an AI model. It can include giving examples, setting a role, or breaking a task into steps.","url":"/glossary/#prompt-engineering"},{"type":"Glossary","title":"Prompt injection","description":"An attack where hidden or malicious instructions, often buried in a web page or document the AI reads, trick the model into ignoring its real instructions. It is a major security concern for AI agents that browse or process outside content.","url":"/glossary/#prompt-injection"},{"type":"Glossary","title":"Quantization","description":"A technique that shrinks a model by storing its numbers with less precision, so it uses less memory and runs faster. It usually costs a little accuracy in exchange for being cheaper to run.","url":"/glossary/#quantization"},{"type":"Glossary","title":"RAG (Retrieval-Augmented Generation)","description":"A method where the system first looks up relevant information from an outside source, such as a document collection, and feeds it to the model so answers are grounded in that material. It helps reduce hallucination and lets a model use up-to-date or private data.","url":"/glossary/#rag-retrieval-augmented-generation"},{"type":"Glossary","title":"Reasoning model","description":"A model trained to spend extra effort thinking through a problem step by step before answering, which improves performance on hard math, coding, and logic tasks. OpenAI's o1 and DeepSeek-R1 are well-known examples.","url":"/glossary/#reasoning-model"},{"type":"Glossary","title":"Red teaming","description":"Deliberately probing an AI system to find its weaknesses, harmful outputs, and ways it can be misused, so they can be fixed before release. The idea and name are borrowed from security testing.","url":"/glossary/#red-teaming"},{"type":"Glossary","title":"Reinforcement learning","description":"A training approach where a system learns by trial and error, receiving rewards for good outcomes and penalties for bad ones, and adjusting to earn more reward. It is used both in game-playing AI and in fine-tuning language models.","url":"/glossary/#reinforcement-learning"},{"type":"Glossary","title":"RLHF (Reinforcement Learning from Human Feedback)","description":"A training method where people rate or rank a model's responses, and those judgments are used to steer the model toward more helpful and appropriate answers. It was central to making chatbots like ChatGPT feel useful and polite.","url":"/glossary/#rlhf-reinforcement-learning-from-human-feedback"},{"type":"Glossary","title":"Scaling laws","description":"Observed, fairly predictable relationships showing that model performance tends to improve as you increase model size, training data, and compute. They have guided the industry's push toward ever-larger models.","url":"/glossary/#scaling-laws"},{"type":"Glossary","title":"Speculative decoding","description":"A speed-up technique where a small, fast model drafts several tokens ahead and a larger model checks them in one pass, accepting the ones it agrees with. It produces the same output as the big model alone but faster.","url":"/glossary/#speculative-decoding"},{"type":"Glossary","title":"Synthetic data","description":"Training data generated by a computer or another AI model, rather than collected from the real world. It can fill gaps where real data is scarce, expensive, or sensitive, though it carries risks if the generated data is flawed.","url":"/glossary/#synthetic-data"},{"type":"Glossary","title":"System prompt","description":"A behind-the-scenes instruction that sets an AI assistant's role, tone, and rules before the conversation with the user begins. Users usually do not see it, but it shapes how the assistant responds.","url":"/glossary/#system-prompt"},{"type":"Glossary","title":"Temperature","description":"A setting that controls how random or predictable a model's output is. Lower values make responses more focused and repeatable; higher values make them more varied and creative.","url":"/glossary/#temperature"},{"type":"Glossary","title":"Test-time compute","description":"The idea of letting a model use more computation while answering, for example by reasoning longer or trying multiple approaches, to get better results. Reasoning models rely heavily on this.","url":"/glossary/#test-time-compute"},{"type":"Glossary","title":"Token","description":"A chunk of text, often a word or part of a word, that a language model reads and generates one at a time. Models measure input length, output length, and pricing in tokens.","url":"/glossary/#token"},{"type":"Glossary","title":"Tool use","description":"When an AI model calls external tools, such as a calculator, a search engine, or a piece of software, to do things it cannot do reliably on its own. It lets models fetch fresh information and take real actions.","url":"/glossary/#tool-use"},{"type":"Glossary","title":"Training data","description":"The collection of examples a model learns from during training. Its size, quality, and biases strongly shape what the model knows and how it behaves.","url":"/glossary/#training-data"},{"type":"Glossary","title":"Training run","description":"A single, complete session of training a model from start to finish on a chosen dataset and setup. For large models, one run can take weeks and cost millions of dollars.","url":"/glossary/#training-run"},{"type":"Glossary","title":"Transformer","description":"The neural network architecture, introduced in 2017, that underlies most modern language models. Its key innovation, the attention mechanism, lets it handle long stretches of text efficiently and learn relationships between distant words.","url":"/glossary/#transformer"},{"type":"Glossary","title":"Turing test","description":"A thought experiment proposed by Alan Turing in 1950, in which a machine passes if a human judge chatting with it cannot reliably tell it apart from a person. It shaped decades of debate about machine intelligence, though many now see it as a limited measure.","url":"/glossary/#turing-test"},{"type":"Glossary","title":"Vector database","description":"A specialized database that stores embeddings and can quickly find the items most similar in meaning to a query. It is a common building block for search and for retrieval-augmented generation.","url":"/glossary/#vector-database"},{"type":"Glossary","title":"Vibe coding","description":"An informal style of programming, popularized in 2025, where a person describes what they want in plain language and lets an AI generate the code, guiding it by feel and results rather than writing much code by hand. The term was coined by AI researcher Andrej Karpathy.","url":"/glossary/#vibe-coding"},{"type":"Glossary","title":"World model","description":"An AI system's internal representation of how the world works, which it can use to predict what happens next or plan ahead. How much today's models truly have one is an open research question.","url":"/glossary/#world-model"},{"type":"Glossary","title":"Zero-shot learning","description":"When a model performs a task it was given no examples for, relying only on its general training and the instructions in the prompt. It contrasts with few-shot, where a few examples are provided.","url":"/glossary/#zero-shot-learning"},{"type":"Coverage","title":"Models news","description":"New model releases, upgrades, and capability jumps — frontier labs and open weights alike.","url":"/topic/models-news/"},{"type":"Coverage","title":"Agents & Tools news","description":"Coding agents, computer use, agent frameworks, and the software people actually build with AI.","url":"/topic/agents/"},{"type":"Coverage","title":"Research news","description":"Papers, benchmarks, training techniques, and measurable progress toward AGI.","url":"/topic/research/"},{"type":"Coverage","title":"Industry news","description":"The labs, the money, the partnerships, and the competition to build general intelligence.","url":"/topic/industry/"},{"type":"Coverage","title":"Safety & Policy news","description":"Alignment, evaluations, governance, and how governments are responding to frontier AI.","url":"/topic/safety/"},{"type":"Coverage","title":"Open Source news","description":"Open-weight models, local inference, and the tooling that puts AI on your own hardware.","url":"/topic/open-source/"},{"type":"Coverage","title":"Hardware & Compute news","description":"GPUs, accelerators, datacenters, and the compute buildout behind the scaling race.","url":"/topic/hardware/"},{"type":"Coverage","title":"Business news","description":"Funding rounds, valuations, enterprise adoption, and AI product economics.","url":"/topic/business/"}]