📚 Dictionary

Here are the explanations for some commonly used terms in the context of AI and Generative Content Creation (AIGC), which can help us become familiar with this field quickly.

English	Chinese	Meaning and Explanation
AI Agents	人工智能代理	Intelligent entities capable of perceiving their environment, making decisions, and taking actions.
Natural Language Processing (NLP)	自然语言处理	A subfield of AI and linguistics focusing on the ability to effectively communicate between humans and computers using natural language. NLP combines methods and technologies from computer science, artificial intelligence, and linguistics to enable computers to understand, interpret, generate, and meaningfully respond to human language.
Large Language Model (LLM)	大语言模型	A shorthand for large-scale language models, referring to models trained with deep learning technology, having a vast number of parameters, and designed for understanding, generating, translating, or performing tasks related to natural language. These models learn the structure, semantics, grammar, and context usage of language through training on extensive text datasets.
Token	-	In large language models, a token represents the smallest unit of meaning that the model can understand and generate, serving as the fundamental unit of large language models.
Prompt	提示词	A text fragment input into an AI model to guide or elicit the model to generate a specific output or complete a specific task.
Prompt Engineering	提示词工程	The optimization of prompt design to enable models to complete specific tasks more effectively or produce higher quality outputs. It includes selecting the right content, format, and context, as well as understanding how to better leverage the model’s capabilities.
Artificial General Intelligence (AGI)	通用型人工智能	There’s no widely accepted definition yet; the following is Microsoft’s definition: AGI is artificial intelligence that has the same capability as humans in any intellectual task.
Generative AI	生成式 AI	A branch of AI that generates new original content based on existing data patterns and examples (such as images, music, or text). Examples include ChatGPT, MidJourney.
Pre-trained	预训练	The initial phase of training a machine learning model, where the model learns general features, patterns, and representations from data without needing to understand the specific tasks it will be applied to later. This unsupervised or semi-supervised learning process allows the model to develop a basic understanding of the underlying data distribution and extract meaningful features that can be used in subsequent fine-tuning for specific tasks.
Transformer	-	The architecture GPT is based on, a deep learning model designed for processing sequential data (such as text). The core of Transformer is the self-attention mechanism, which can handle long-distance dependencies and improve the model’s parallel processing capabilities.
Generative Pre-trained Transformer (GPT)	生成式预训练模型	A large language model based on the Transformer architecture, developed by OpenAI. It is an advanced natural language processing technology capable of generating coherent and highly relevant text, performing various language tasks such as text generation, translation, question answering, and summarization.
Contrastive Language–Image Pretraining (CLIP)	-	An AI model developed by OpenAI that connects images and text, enabling it to understand and generate descriptions of images.
Foundation Model/Checkpoint	基础模型	Refers to large, pre-trained models trained on a wide range of datasets to capture rich data representations and capable of being fine-tuned or adapted for use on many different tasks.
Machine Learning	机器学习	A field of artificial intelligence science, focusing on how to improve the performance of specific algorithms through experiential learning. It provides systems the ability to automatically learn and improve from experience without explicit programming.
Deep Learning	深度学习	A subset of machine learning that uses algorithms known as neural networks for learning and decision-making. Deep learning simulates the way the human brain works by processing data through multi-layer (i.e., “deep”) neural network structures, enabling computers to learn from experience and recognize patterns.
Supervised Learning	监督学习	A type of machine learning where the model is trained using a training dataset comprising input data and corresponding labels (or targets). During supervised learning, the model attempts to learn the mapping between input data and output labels so that it can predict the correct output when given new, unseen input data.
Unsupervised Learning	无监督学习	A type of machine learning that, unlike supervised learning, does not rely on labeled data for learning. In unsupervised learning, algorithms are given a large amount of data without additional labels or indications of what the output should be. The goal is for the algorithm to discover patterns, structures, or knowledge within the data on its own.
Reinforcement Learning	强化学习	A type of machine learning that focuses on learning how to achieve a goal in a specific environment through trial and error. In reinforcement learning, an agent learns to choose actions that maximize some cumulative reward by interacting with the environment. This learning process involves evaluating which actions will bring the greatest benefit in the long term, not just immediate rewards.
Transfer Learning	迁移学习	A type of machine learning that allows a model to apply knowledge learned in one task to another related task. The core idea of this approach is to use the knowledge gained from previous tasks to accelerate or improve the learning process for subsequent tasks, especially when the subsequent task has less data. Transfer learning is particularly significant in the field of deep learning because deep learning models typically require a lot of data and computational resources to train from scratch.
Inference	推理	The process of making predictions using a trained machine learning model.
Embedding	向量化	A key technology for enhancing data processing and numerical computation performance, especially applicable in fields such as scientific computing, machine learning, deep learning, etc. It involves reducing loops and utilizing underlying hardware optimizations to make code execution faster and more efficient.
Parameters	参数	In machine learning, parameters are internal variables used by the model to make predictions. They are learned from the training data during the training process. For example, weights and biases are parameters in a neural network.
Fine-tuning	微调	Refers to further training a model that has already been trained, on a specific task, to adapt the model to the specific data and requirements of that task. During fine-tuning, the model’s parameters are further adjusted using a smaller, task-specific dataset to learn task-specific patterns and improve performance on the new task.
Chain of Thought	思维链	An important technique in large model prompt engineering, often used to describe the sequence of reasoning steps an AI model uses to reach a decision.
Multimodal	多模态	Modality refers to the ways things are experienced and occur. We live in a world composed of various modal information, including visual, auditory, textual, olfactory, etc. In the field of artificial intelligence, multimodal typically refers to information from multiple modalities, including text, images, videos, audio, etc.
Hallucination	幻觉	Commonly used to describe outputs generated by models or algorithms that do not align with reality.