AI 问题，清晰作答
用直白的语言解释人工智能、机器学习、神经网络和大语言模型。理解概念、查看真实示例，并通过交互演示亲手实验。
12 个问题
AI 基础阅读约 5 分钟
What Is a Neural Network?
What a neuron computes, why a hidden layer needs a non-linearity, and a worked example you can check on paper.
机器学习阅读约 4 分钟
How Does Backpropagation Work?
The chain rule as a credit-assignment algorithm, with a one-weight example you can check by hand and the role of the learning rate.
机器学习阅读约 5 分钟
What Is an Embedding in AI?
How meaning becomes geometry, what cosine similarity measures, and why embeddings are the retrieval layer behind modern AI products.
深度学习阅读约 4 分钟
What Is Self-Attention in Transformers?
How tokens decide what to read, the scaled dot-product formula, causal masking and why long context costs quadratic compute.
AI 基础阅读约 4 分钟
What Is the Difference Between Training and Inference?
What changes in each phase, why inference is cheaper per request, and why chatting with a model does not retrain it.
机器学习阅读约 4 分钟
What Is a Loss Function in Machine Learning?
What a loss rewards, how MSE and cross-entropy differ, and why a low training loss is not the same as a good model.
AI 基础阅读约 4 分钟
What Are Parameters in an AI Model?
What counts as a parameter, a layer-by-layer count you can verify, and what the number does — and does not — tell you.
大语言模型阅读约 4 分钟
Why Do AI Models Hallucinate?
Why plausible text is not verified text, how grounding and abstention help, and why hallucination cannot be eliminated entirely.
大语言模型阅读约 4 分钟
What Does Temperature Do in an LLM?
The softmax formula behind the dial, what low and high values actually do, and why temperature cannot add knowledge.
大语言模型阅读约 4 分钟
What Is Top-p Sampling?
How the nucleus is chosen, a worked probability example, and how top-p interacts with temperature and top-k.
大语言模型阅读约 4 分钟
Why Does ChatGPT Give Different Answers to the Same Prompt?
Sampling as the main cause, the other four sources of variation, and the settings that make outputs reproducible enough to test.
AI 基础阅读约 4 分钟
What Is the Difference Between AI, Machine Learning and Deep Learning?
The nested relationship between the three terms, a concrete example at each level, and why the distinction changes your engineering choices.