Approved 2025/05/28 16:55 by ddehamer (version: 1) | Approver: @ai-us-principals
Key Terms for AI
| Term / Technology | Category | Definition / Description |
| Artificial Intelligence (AI) | Concept | Simulation of human intelligence by machines, especially computer systems. |
| Machine Learning (ML) | Concept | Algorithms that allow computers to learn patterns and make decisions with data. |
| Deep Learning | Subfield of ML | Uses neural networks with many layers to model complex representations. |
| Natural Language Processing (NLP) | Subfield of AI | Enables machines to understand and generate human language. |
| Computer Vision | Subfield of AI | AI techniques to interpret images and videos. |
| Reinforcement Learning | ML Methodology | Models learn by interacting with an environment and receiving feedback. |
| Supervised Learning | ML Type | Model is trained on labeled input-output pairs. |
| Unsupervised Learning | ML Type | Model learns from unlabeled data by identifying patterns. |
| Transformer | Model Architecture | Deep learning architecture using self-attention, core to GPT and BERT models. |
| Token | NLP Concept | A unit of text, like a word or subword, used in language models. |
| Prompt Engineering | AI Interaction Design | Crafting model inputs to steer the output toward desired results. |
| Inference | AI Operation | Running data through a trained model to get predictions or outputs. |
| Training | AI Operation | Teaching a model using data so it can make accurate predictions. |
| Bias | Ethical Concern | Systematic errors in AI that can lead to unfair or discriminatory results. |
| Explainability | AI Governance | Making AI models' decisions interpretable and understandable to humans. |
| Tuning / Fine-tuning | Model Optimization | Adjusting a pre-trained model for a specific task or domain. |
| Overfitting | ML Issue | When a model performs well on training data but poorly on new data. |
| Model Context Protocol (MCP) | Model Integration | Standardized interface or protocol designed to allow language models (LLMs) to access, interpret, and maintain context across multiple interactions, tools, and data sources. |
| Technology | Category | Description |
| OpenAI | Foundation Model Provider | Creator of ChatGPT and GPT models. Offers APIs for LLMs, embeddings, and other AI services. |
| Azure OpenAI Service | Cloud Platform | Microsoft’s hosted version of OpenAI models with enterprise security, scaling, and governance. |
| Anthropic | Foundation Model Provider | AI company focused on safety and alignment, creator of the Claude model family. |
| Mistral (MCP) | Foundation Model Provider | European AI company building open-weight language models, known for fast inference and performance. |
| Agentic AI | AI Paradigm | AI systems that act autonomously and can make decisions, take actions, and pursue goals across tools and environments. |
| LangChain | Agentic Framework | Framework to build agent-based AI applications using chains of prompts, tools, and models. |
| AutoGPT | Agentic Framework | Autonomous AI agent that breaks down tasks into subtasks and self-prompt to complete goals. |
| Hugging Face | AI Platform / Model Hub | Open-source platform hosting thousands of models, datasets, and transformers tools. |
| LLMOps | Operational Practice | Managing, deploying, monitoring, and improving LLMs in production (similar to MLOps). |
| Retrieval-Augmented Generation (RAG) | AI Technique | Combines LLMs with external knowledge sources (e.g., vector databases) for more accurate and current answers. |
AI Knowledge
