学习
学习
关于复杂 AI 主题的解读、入门与背景。
学习路径
精选知识库阅读顺序。
- New to AI? Start here
A short path from the basics of AI through chatbots and large language models.
companies
concepts
- beginnerWhat is artificial intelligence?Artificial intelligence is a family of computer systems that perform tasks people associate with human intelligence — recognizing patterns, generating language, or recommending actions — usually by learning statistical relationships from data rather than following only hand-written rules.
- beginnerWhat is ChatGPT?ChatGPT is OpenAI’s consumer and business chatbot product. It wraps large language models in a conversational interface with memory, tools, and subscription tiers — making generative AI mainstream while remaining a product layer on top of rapidly changing models.
models
- beginnerWhat is a large language model?A large language model is a neural network trained on vast amounts of text to predict the next token in a sequence. That simple objective yields systems that can answer questions, write code, and follow instructions — with uneven reliability and no built-in understanding of truth.
- intermediateWhat are model weights?Model weights are the numerical parameters a neural network learns during training. Releasing weights lets others run or fine-tune a model locally; keeping them private forces users through an API. Licensing and leak stories turn “weights” into a business and policy flashpoint.
- intermediateWhat is a context window?A context window is the maximum amount of text (measured in tokens) a model can consider at once — the prompt, conversation history, and any retrieved documents. Larger windows enable longer briefs and codebases in one shot, but cost and attention quality do not scale linearly.
- advancedWhat is mixture of experts?Mixture of experts (MoE) is an architecture where many specialist “expert” networks sit behind a router that activates only a subset per token. That lets models grow total parameter count while keeping per-token compute closer to a smaller dense model — with routing and load-balancing tradeoffs.
training
- intermediateWhat is fine-tuning?Fine-tuning continues training a pretrained model on a narrower dataset so it adopts a style, domain, or task. It is more invasive than prompting alone, cheaper than training from scratch, and a common way enterprises specialize foundation models without owning pretraining.
- intermediateWhat is RAG?Retrieval-augmented generation (RAG) is a pattern where a system searches a knowledge base for relevant passages, then conditions the language model on those passages when answering. It aims to ground responses in your documents without permanently baking every fact into model weights.
- advancedWhat is RLHF?Reinforcement learning from human feedback (RLHF) trains a model to prefer outputs that human raters rank higher, usually after supervised fine-tuning. It is a major reason chat models sound helpful and refuse some requests — and a source of debates about whose preferences get encoded.