TechCrunch presents an extensive, regularly updated glossary designed to help readers understand the rapidly evolving vocabulary of artificial intelligence.

As AI becomes increasingly embedded in technology, business, and everyday tools, the industry has developed a dense set of specialized terms that can be difficult to follow.

This glossary aims to make those concepts accessible to developers, investors, and general readers who encounter AI-related discussions in media, product meetings, and technical conversations.The article defines a wide range of foundational and advanced AI concepts.

It begins with major ideas such as artificial general intelligence (AGI), describing competing interpretations from organizations like OpenAI and Google DeepMind, and moves into applied system concepts such as AI agents, which are tools capable of performing multi-step tasks autonomously.

It also explains infrastructure-related terms like API endpoints, compute resources (GPUs, TPUs, CPUs), and inference, which refers to the process of running trained models.Core machine learning techniques are also covered, including deep learning, neural networks, reinforcement learning, and fine-tuning.

The glossary further explains generative AI mechanisms such as diffusion models and GANs, along with optimization techniques like distillation and parallelization.

Additional modern industry concepts include large language models (LLMs), tokenization, memory caching, mixture of experts architectures, and emerging standards like the Model Context Protocol (MCP).

The glossary also addresses challenges and risks, including hallucinations—when models generate incorrect information—and systemic issues such as RAM shortages driven by AI infrastructure demand.

Overall, the article serves as a living reference guide to help readers navigate the fast-changing AI landscape by clarifying technical terminology and contextualizing how these systems are built, trained, and deployed.