Back to Blog
April 12, 2026 • Natasha Lomas, Romain Dillet, Kyle Wiggers, Lucas Ropek

AI Glossary: Understanding Key Terms

AI Glossary: Understanding Key Terms AI Glossary: Understanding Key Terms AI Glossary: Understanding Key Terms

AI Glossary: Understanding Key Terms

Artificial intelligence is a rapidly evolving field, often characterized by complex jargon. This glossary aims to demystify common AI terms, providing clear definitions to aid understanding.

Key AI Concepts

  • Artificial General Intelligence (AGI): AI that possesses human-level cognitive capabilities across a wide range of tasks. Definitions vary among experts, but it generally implies AI that can perform most economically valuable work as well as or better than humans.
  • AI Agent: An AI system designed to perform a series of tasks autonomously on behalf of a user. This goes beyond simple chatbots and can involve complex actions like managing expenses or writing code. The infrastructure for advanced AI agents is still under development.
  • Chain of Thought (CoT): A reasoning technique for Large Language Models (LLMs) that involves breaking down complex problems into smaller, intermediate steps. This improves the accuracy and reliability of the AI's output, especially for logic and coding tasks.
  • Compute: Refers to the computational power required to train and run AI models. This includes hardware like GPUs, CPUs, and TPUs, forming the essential infrastructure for the AI industry.
  • Deep Learning: A subset of machine learning that uses multi-layered artificial neural networks, inspired by the human brain. Deep learning models can identify complex patterns in data and learn from errors, but they require vast amounts of data and significant training time.
  • Diffusion: A generative AI technique, often used in image and music generation, that works by gradually adding noise to data and then learning to reverse the process to create new content. It's inspired by physical diffusion processes.
  • Distillation: A method of knowledge transfer where a smaller "student" model is trained to mimic the behavior and outputs of a larger "teacher" model. This can create more efficient models.
  • Fine-tuning: The process of further training a pre-existing AI model on a specialized dataset to optimize its performance for a specific task or domain.
  • Generative Adversarial Network (GAN): A framework involving two competing neural networks (a generator and a discriminator) used to create realistic synthetic data, such as deepfakes. The adversarial nature helps improve the realism of the generated output.
  • Hallucination: The phenomenon where AI models generate incorrect or fabricated information. This is a significant challenge for AI quality and can lead to misinformation. It's often attributed to gaps in training data and drives the development of more specialized AI models.
  • Inference: The process of running a trained AI model to make predictions or generate outputs based on new data. The efficiency of inference depends heavily on the hardware used.
  • Large Language Model (LLM): AI models, like those powering ChatGPT or Gemini, that are trained on vast amounts of text data to understand and generate human-like language. They work by predicting the most probable sequence of words based on the input prompt.
  • Memory Cache: An optimization technique that stores the results of previous computations to speed up future inference requests. KV caching is a common example in transformer-based models.
  • Neural Network: The foundational algorithmic structure for deep learning and many modern AI systems. Inspired by the brain's structure, neural networks leverage advancements in hardware (like GPUs) to achieve high performance.
  • RAMageddon: A term describing the current shortage and increasing cost of Random Access Memory (RAM) chips, driven by high demand from the AI industry, impacting various tech sectors.
  • Training: The process of feeding data into an AI model to enable it to learn patterns and achieve a desired outcome. This is a computationally intensive process, though hybrid approaches like fine-tuning can reduce costs.
  • Tokens: The basic units of data processed by LLMs. Tokenization breaks down human language into segments digestible by AI. Token usage is often the basis for AI service pricing.
  • Transfer Learning: A technique where a model trained for one task is repurposed as a starting point for a related task. This can save development time and resources, especially when data is limited.
  • Weights: Numerical parameters within an AI model that determine the importance of different input features during training. They are adjusted during the training process to optimize the model's output.