The only AI glossary you need this year

Artificial intelligence is rewriting the world, and simultaneously inventing a whole new language to describe how it is doing it. Sit in on any product meeting, pitch, or panel these days, and you will hear people toss around LLMs, RAG, RLHF, and a dozen other terms that can make even very smart people in the tech world feel a little insecure. This glossary is our attempt to fix that: plain-English definitions of the AI terms you are most likely to actually run into, whether you are building with this stuff, investing in it, or just trying to keep up by reading Bitcoin World or listening to related podcasts. We update it regularly as the field evolves, so consider it a living document, much like the AI systems it describes.

Core AI concepts: from AGI to neural networks

Understanding AI starts with a few foundational terms. AGI, or artificial general intelligence, refers to AI that is more capable than the average human at many tasks — though definitions vary. OpenAI CEO Sam Altman once described AGI as the “equivalent of a median human that you could hire as a co-worker,” while Google DeepMind views it as “AI that is at least as capable as humans at most cognitive tasks.” Neural networks are the multi-layered algorithmic structures that underpin deep learning, inspired by the interconnected pathways of the human brain. The rise of graphical processing hardware (GPUs) unlocked the power of this theory, enabling neural network-based AI systems to achieve far better performance across voice recognition, autonomous navigation, and drug discovery.

Deep learning is a subset of machine learning where AI algorithms are designed with multi-layered neural networks, allowing them to make complex correlations. Deep learning models can identify important characteristics in data themselves, but they require millions of data points and take longer to train than simpler algorithms. Large language models (LLMs) are the AI models used by popular assistants like ChatGPT, Claude, and Gemini. They are deep neural networks made of billions of numerical parameters that learn the relationships between words and phrases, creating a multidimensional map of language.

How AI models work: training, inference, and reasoning

Training is the process of feeding data into a model so it can learn patterns and generate useful outputs. It can be expensive because it requires lots of inputs, and volumes have been trending upward. Inference is the process of running an AI model — setting it loose to make predictions or draw conclusions from previously seen data. Inference cannot happen without training; a model must learn patterns before it can effectively extrapolate.

Chain-of-thought reasoning means breaking down a problem into smaller, intermediate steps to improve the quality of the end result. It usually takes longer to get an answer, but the answer is more likely to be correct, especially in logic or coding contexts. Reasoning models are developed from traditional LLMs and optimized for chain-of-thought thinking using reinforcement learning. Reinforcement learning is a way of training AI where a system learns by trying things and receiving rewards for correct answers — like training a pet with treats, except the “treat” is a mathematical signal indicating success. Techniques like reinforcement learning from human feedback (RLHF) are central to how leading AI labs fine-tune their models.

Tokens are the basic building blocks of human-AI communication, representing discrete segments of data processed by an LLM. In enterprise settings, tokens determine cost — most AI companies charge for LLM usage on a per-token basis. Token throughput measures how much AI work a system can handle at once, determining how many users a model can serve simultaneously and how quickly each receives a response.

Specialized architectures and techniques

Mixture of Experts (MoE) is a model architecture that splits a neural network into many smaller specialized sub-networks, or “experts,” and only activates a handful for any given task. This makes it possible to build enormous models that stay relatively fast and cheap to run. Mistral AI’s Mixtral model is a well-known example. Diffusion is the tech at the heart of many art-, music-, and text-generating AI models. Inspired by physics, diffusion systems slowly “destroy” the structure of data by adding noise until nothing is left, then learn a reverse process to restore the data from noise.

Distillation is a technique used to extract knowledge from a large AI model using a teacher-student model. Developers send requests to a teacher model and record the outputs, then use these outputs to train a smaller student model. This is likely how OpenAI developed GPT-4 Turbo. Fine-tuning refers to further training an AI model to optimize performance for a more specific task by feeding in new, specialized data. Many AI startups take LLMs as a starting point and fine-tune them with domain-specific knowledge.

GANs, or Generative Adversarial Networks, involve a pair of neural networks — one generates outputs while the other evaluates them. The structured contest can optimize AI outputs to be more realistic without additional human intervention, though GANs work best for narrower applications like producing realistic photos or videos.

Practical applications: agents, coding, and protocols

AI agents are tools that use AI technologies to perform a series of tasks on your behalf — beyond what a basic AI chatbot could do — such as filing expenses, booking tickets, or writing and maintaining code. The basic concept implies an autonomous system that may draw on multiple AI systems to carry out multistep tasks. Coding agents are specialized versions applied to software development. Rather than simply suggesting code for a human to review, a coding agent can write, test, and debug code autonomously, handling the iterative, trial-and-error work that typically consumes a developer’s day.

Model Context Protocol (MCP) is an open standard that lets AI models connect to outside tools and data — your files, databases, or apps like Slack and Google Drive — without a developer building a custom connector for every pairing. Anthropic introduced MCP in 2024 and later handed it over to the Linux Foundation, and it has since been adopted by OpenAI, Google, and Microsoft.

Key challenges: hallucination, overfitting, and memory shortages

Hallucination is the AI industry’s preferred term for models making stuff up — literally generating information that is incorrect. Hallucinations produce outputs that can be misleading and could lead to real-life risks, such as harmful medical advice. The problem is thought to arise from gaps in training data and is contributing to a push toward increasingly specialized AI models. Validation loss is a number that tells you how well an AI model is learning during training — lower is better. It helps flag overfitting, a condition in which a model memorizes its training data rather than truly learning patterns it can generalize to new situations.

RAMageddon is the term for an ever-increasing shortage of RAM chips, which power pretty much all tech products. As the AI industry has blossomed, the biggest tech companies are buying so much RAM to power their data centers that there is not much left for the rest of us. This supply bottleneck affects gaming, consumer electronics, and general enterprise computing, with prices expected to stay high until the shortage ends.

Conclusion

This glossary covers the most essential AI terms you are likely to encounter in 2025, from foundational concepts like neural networks and LLMs to emerging standards like MCP and challenges like RAMageddon. The field evolves rapidly, and definitions can shift as technology advances. We update this guide regularly to reflect new developments, so bookmark it and check back as the AI landscape continues to transform.

FAQs

Q1: What is the difference between AGI and a regular AI model?
AGI refers to artificial general intelligence — AI that is more capable than the average human at many tasks. Current AI models, including LLMs, are narrow AI systems designed for specific tasks like text generation or image recognition. AGI remains a theoretical goal that has not yet been achieved.

Q2: Why do AI companies charge based on tokens?
Tokens are the basic units of text that AI language models process. Most AI companies charge for LLM usage on a per-token basis because it directly reflects the computational resources required to process a request. The more tokens a business uses, the more it pays.

Q3: What is the difference between open source and closed source AI models?
Open source AI models have their underlying code publicly available for anyone to use, inspect, or modify — Meta’s Llama family is a prominent example. Closed source models keep the code private, so you can use the product but not see how it works, as with OpenAI’s GPT models. This distinction has become one of the defining debates in the AI industry.