Wiki
Hand-curated AI glossary: concepts, terms, models, products. Cross-linked, no UGC.
AI Concepts
14-
Constitutional AI (CAI)
Constitutional AI (CAI) is a training paradigm Anthropic proposed in 2022, aiming to repla…
-
In-Context Learning (ICL)
In-context learning (ICL) is one of the most important capabilities of LLMs: putting a few…
-
Mechanistic Interpretability
Mechanistic interpretability (mech interp) is a branch of AI safety and alignment research…
-
Model Calibration
Calibration measures whether a model's self-reported confidence actually matches its a…
-
RLVR (Reinforcement Learning with Verifiable Rewards)
RLVR is the new RL paradigm for large models that emerged from 2024 onward: the reward sig…
-
Scaling Laws
Scaling laws describe the power-law connection between large models' parameter count N…
-
Self-Play
Self-play is a reinforcement-learning training paradigm: agents play against / collaborate…
-
Speculative Decoding
Speculative decoding is a popular LLM inference acceleration technique that emerged from 2…
-
Structured Output
Structured output is the capability to make an LLM emit values that conform to a predefine…
-
Synthetic Data
Synthetic data is training data created by a model (usually a stronger model or a speciali…
-
System 1 / System 2 (Dual-Process Thinking)
System 1 / System 2 is the cognitive framework psychologist Daniel Kahneman laid out in Th…
-
Tool Use
Tool use is the capability that lets an LLM, on its own initiative, decide to call externa…
-
Transformer Architecture
The Transformer is a neural network architecture introduced by Google in the 2017 paper At…
-
World Model
A world model is a visionary concept in AI research: building an internal representation o…
Concepts
41-
AGI (Artificial General Intelligence)
AGI (Artificial General Intelligence) refers to artificial intelligence that matches or ex…
-
AI Agent
An AI Agent autonomously perceives its environment, makes decisions, and takes actions. LL…
-
AI Alignment
AI Alignment is the field of research that makes AI systems' behavior consistent with …
-
AI Search (Answer Engine)
AI Search (Answer Engine) refers to search products that use an LLM to directly generate a…
-
BLEU / ROUGE
BLEU and ROUGE are the most classic automatic metrics for NLP text generation tasks. Still…
-
BM25
BM25 (Best Matching 25) is a classic information retrieval ranking algorithm, based on ter…
-
Benchmark
A Benchmark is a standard test set used to objectively measure model performance on a spec…
-
CUDA
CUDA is NVIDIA's GPU parallel computing API. Deep learning depends almost entirely on …
-
Chain-of-Thought
Chain-of-Thought (CoT) is the prompting technique that makes an LLM explicitly output inte…
-
Code Generation
Code Generation is the capability of an LLM to write runnable code from natural language d…
-
Context Window
The context window is the maximum number of tokens an LLM can "see" in one go. Con…
-
DPO
DPO (Direct Preference Optimization) is a simplified version of RLHF : train directly on p…
-
Embedding
Embedding maps arbitrary data (text, image, audio, user behavior) into a fixed-dimensional…
-
Fine-tuning
Fine-tuning continues training a pre-trained LLM on a specific dataset to adapt it to a ta…
-
FlashAttention
FlashAttention is the attention algorithm rewrite proposed by Tri Dao et al. in 2022: keep…
-
Function Calling
Function Calling lets an LLM output structured JSON specifying which tool to invoke and wi…
-
Hallucination
Hallucination is when an LLM generates content that looks plausible but is actually wrong …
-
KV Cache
KV Cache (Key-Value Cache) is the key optimization during inference : cache the Key and Va…
-
Large Language Model
A Large Language Model (LLM) is a Transformer -based language model pre-trained on massive…
-
LoRA
LoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning (PEFT) method: freeze the …
-
MCP (Model Context Protocol)
MCP (Model Context Protocol) is an open protocol released by Anthropic at the end of 2024,…
-
Mamba
Mamba is a new-generation sequence modeling architecture proposed in 2023 that uses Select…
-
MoE (Mixture of Experts)
MoE (Mixture of Experts) is a "sparse activation for larger parameter count" model…
-
Model Distillation
Model Distillation is the process of "distilling" knowledge from a large model (te…
-
Model Inference
Inference is the process of using a trained model in production (as opposed to training or…
-
Model Quantization
Quantization converts model weights from high precision (FP32/FP16) to lower precision (IN…
-
Multimodal Models
Multimodal models process multiple data types simultaneously: text, images, audio, video. …
-
PPO (Proximal Policy Optimization)
PPO (Proximal Policy Optimization) is the reinforcement learning algorithm proposed by Ope…
-
Perplexity
Perplexity (PPL) is the most fundamental metric for evaluating language models. It measure…
-
Pre-training
Pre-training is the first stage of LLM training: using massive unlabeled text (trillion-to…
-
Prompt Engineering
Prompt Engineering is the practice of designing the input to an LLM to get desired outputs…
-
RAG (Retrieval-Augmented Generation)
Retrieval-Augmented Generation (RAG) is a paradigm that connects external knowledge bases …
-
RLHF
RLHF (Reinforcement Learning from Human Feedback) is the advanced form of fine-tuning : us…
-
Reasoning Model
Reasoning Models are LLMs specifically optimized for "slow thinking" tasks: before…
-
Rerank
Rerank is the second step of the RAG retrieval flow: use a stronger model to refine first-…
-
SWE-bench
SWE-bench (Software Engineering Benchmark) is the most authoritative code engineering benc…
-
Self-Attention
Self-Attention is the core computational unit of the Transformer architecture . It lets ev…
-
Structured Output (JSON Mode)
Structured output means making the LLM output strictly conforming to a predefined schema (…
-
Supervised Fine-Tuning
Supervised Fine-Tuning (SFT) is the most common form of fine-tuning : train on (instructio…
-
Token
A token is the smallest unit of text that a large language model processes. One token roug…
-
Tokenization
Tokenization is the process of splitting raw text into the tokens a model can understand. …
Infrastructure
9-
Docker
Docker is the de facto standard containerization platform. Packages applications
-
Elasticsearch
Elasticsearch (ES) is the distributed search engine built on Apache Lucene, the most popul…
-
Hugging Face
Hugging Face (huggingface.co) is the "GitHub + AWS" of AI: open-source model hosti…
-
LangChain
LangChain is the most popular LLM application development framework, covering the full pip…
-
LlamaIndex
LlamaIndex is a framework focused on connecting LLMs with private data, the mainstream cho…
-
Ollama
Ollama is the simplest tool for running open-source LLMs locally or at the edge. One comma…
-
Vector Database
A vector database is a database system purpose-built for storing and retrieving high-dimen…
-
pgvector
pgvector is the open-source vector retrieval extension for PostgreSQL (GitHub 5k+ stars). …
-
vLLM
vLLM is the open-source high-throughput LLM inference engine from UC Berkeley RISELab. The…
Models & Products
23-
Anthropic
Anthropic is a frontier AI safety company founded in 2021 by former OpenAI VP of Research …
-
ChatGPT
ChatGPT is OpenAI's conversational AI product launched in November 2022, the milestone…
-
Claude
Claude is Anthropic's large language model family, known for safety, reasoning ability…
-
Cohere
Cohere is a Canadian AI company (founded 2020), focused on enterprise-grade LLMs and RAG t…
-
Cursor
Cursor is the AI-first code editor from Anysphere, based on a VS Code fork. In 2024-2025 i…
-
DeepSeek
DeepSeek is the large language model family released by DeepSeek AI. In 2024-2025, V3 / R1…
-
Fireworks AI
Fireworks AI is a US-based frontier-model inference platform founded in 2022, focused on s…
-
GPT-5
GPT-5 is OpenAI's flagship large language model released in 2025. Compared to GPT-4, i…
-
Gemini
Gemini is Google DeepMind's large language model family released at end of 2023, Googl…
-
Google DeepMind
Google DeepMind is Google's AI research lab, descended from DeepMind Technologies, fou…
-
Groq
Groq is an American AI inference chip + cloud service company. Its self-developed LPU (Lan…
-
Llama
Llama is Meta's open-source large language model family, currently the benchmark for o…
-
Meta AI
Meta AI is the AI division of Meta (Facebook's parent company), tracing back to FAIR (…
-
MiniMax
MiniMax (Shanghai Xiyu Technology / 稀宇科技) is a leading Chinese multimodal foundation-model…
-
Mistral AI
Mistral AI is a Paris, France AI company (founded 2023), focused on open-source LLMs and i…
-
NVIDIA
NVIDIA is the GPU and accelerated-computing company founded in 1993 by Jensen Huang (黄仁勋).…
-
OpenRouter
OpenRouter (openrouter.ai) is an AI model API aggregation platform. It unifies models from…
-
Qwen (Tongyi Qianwen)
Qwen (通义千问) is Alibaba Cloud's LLM family, with Chinese capability among the strongest…
-
Stability AI
Stability AI is a London-based open-source generative AI company founded in 2019 by Emad M…
-
Together AI
Together AI is a San Francisco-based frontier-model cloud platform founded in 2022 with th…
-
TypeSafe AI
TypeSafe AI is an AI lab that came out of stealth on 2026-09-15, founded by former OpenAI …
-
Whisper
Whisper is OpenAI's open-source automatic speech recognition (ASR) model released in 2…
-
xAI
xAI is the AI lab founded by Elon Musk in 2023 with the stated mission of "understandi…
Security
4-
Data Poisoning
Data Poisoning means attackers inject malicious samples into pretraining / SFT / RLHF data…
-
Jailbreak
Jailbreak means using carefully crafted prompts to bypass LLM safety alignment, making the…
-
Prompt Injection
Prompt Injection is the most serious security threat to LLM applications: attackers embed …
-
Red Team
Red Team means organizing specialized teams (internal or external) to simulate attackers, …
没有匹配的词条 — 换个关键词试试。