AI Glossary Key Terms Explained From LLMs to Hallucinations

Artificial intelligence is a complex field filled with technical jargon. To help demystify the terminology, we've compiled a glossary of essential AI terms used in industry coverage. This guide will be regularly updated as AI research evolves.

AGI (Artificial General Intelligence)

AGI refers to AI that outperforms humans in most cognitive tasks. Definitions vary: OpenAI's Sam Altman describes it as a "median human co-worker," while Google DeepMind sees it as AI matching human capabilities. Experts still debate its exact meaning. Learn more about AGI confusion.

AI Agent

An AI agent performs multi-step tasks autonomously, like booking tickets or managing expenses. While promising, infrastructure is still developing. More on AI agents.

Chain of Thought

This reasoning method breaks problems into smaller steps for more accurate results, especially in logic or coding contexts.

Deep Learning

A machine learning subset using multi-layered neural networks to identify complex patterns. Requires massive datasets but delivers superior performance.

Diffusion

The technology behind many generative AI models, inspired by physics. Systems learn to reconstruct data from noise.

Distillation

A technique to create smaller, efficient models by mimicking larger "teacher" models. Used in developing GPT-4 Turbo. Microsoft's investigation into improper use.

Fine-Tuning

Specializing pre-trained models for specific tasks by adding domain-specific data.

GAN (Generative Adversarial Network)

A framework where two neural networks compete to generate realistic outputs, used in deepfake technology.

Hallucination

AI's tendency to fabricate information - a major quality challenge. Results from training data gaps and contributes to specialized AI development.

Inference

The process of using a trained AI model to make predictions. Requires appropriate hardware for optimal performance.

Large Language Model (LLM)

The foundation of AI assistants like ChatGPT and Gemini. These neural networks with billions of parameters learn language patterns from vast text datasets.

Neural Network

The interconnected algorithmic structure inspired by human brains that powers modern AI.

Training

The data-driven process that transforms random numbers into functional AI models. Can be resource-intensive.

Transfer Learning

Adapting pre-trained models for new tasks, saving development resources.

Weights

Numerical parameters determining feature importance in AI models, adjusted during training to improve accuracy.

This glossary will continue expanding as AI technology advances.