The only AI glossary you’ll need this year

The field of artificial intelligence is rapidly evolving, introducing a new lexicon that can be challenging to navigate. Terms like LLMs, RAG, RLHF, and others are frequently used in product meetings, pitches, and panels, often leaving even tech-savvy individuals feeling uncertain. This glossary aims to address this by providing clear, understandable definitions for the most common AI terms encountered today, whether you are developing, investing, or simply keeping up with the latest advancements. It is a living document, regularly updated to reflect the dynamic nature of AI.

Artificial General Intelligence (AGI) refers to AI systems that are exceptionally capable, often surpassing human performance across various tasks. While definitions vary among leading AI organizations like OpenAI and Google DeepMind, the core concept revolves around highly autonomous systems that can perform most economically valuable work at a human or superhuman level. The precise understanding of AGI continues to be a subject of ongoing debate and research within the AI community.

AI agents are tools that utilize AI technologies to autonomously perform a series of tasks, exceeding the capabilities of basic chatbots. These agents can handle complex functions such as expense filing, booking appointments, or even writing and maintaining code. While the exact definition can vary, the fundamental idea is an autonomous system capable of drawing on multiple AI systems to execute multi-step tasks.

API endpoints serve as crucial "buttons" that allow different software programs to interact and exchange data. Developers leverage these interfaces to build integrations, enabling applications to pull data from each other or allowing AI agents to control third-party services directly. As AI agents become more sophisticated, they are increasingly able to discover and utilize these endpoints independently, leading to powerful and sometimes unexpected automation capabilities across various platforms and smart devices.

Chain-of-thought reasoning in large language models involves breaking down complex problems into smaller, intermediate steps. This method is utilized to enhance the accuracy and quality of the final output, particularly in logical or coding contexts. While it may extend the time required to arrive at an answer, this approach significantly improves correctness. These reasoning models are developed from traditional LLMs and optimized for this type of thinking through reinforcement learning.

Coding agents are specialized AI programs designed for software development that can autonomously write, test, and debug code. Unlike tools that merely suggest code for human review, coding agents can independently handle iterative tasks and trial-and-error processes, operating across entire codebases. They can identify bugs, run tests, and push fixes with minimal human oversight, acting as highly efficient, tireless assistants to developers, though human review remains essential.

"Compute" refers to the essential computational power that enables AI models to operate, fueling the entire AI industry by facilitating the training and deployment of powerful models. This term often encompasses the hardware infrastructure that provides this power, including GPUs, CPUs, and TPUs, which form the foundational bedrock of modern AI.

Deep learning is a subset of machine learning that employs multi-layered artificial neural networks (ANNs). This architecture allows deep learning algorithms to identify more complex correlations in data compared to simpler machine learning systems. Inspired by the human brain's neural pathways, these models can learn from errors and improve through repetition, though they require vast amounts of data and significant computational resources for training.