Artificial intelligence terminology continues to expand as researchers and companies develop new systems, prompting the need for clearer definitions of commonly used concepts. A glossary compiled to explain these terms outlines how core ideas—from artificial general intelligence to tokens—are used across the industry, while noting that some definitions remain debated among experts.
Artificial General Intelligence And AI Agents
Artificial general intelligence, often shortened to AGI, refers broadly to systems that can perform tasks at or beyond the level of an average human across many domains. Sam Altman has described AGI as equivalent to a median human worker, while OpenAI defines it as systems that outperform humans in most economically valuable work. Google DeepMind describes AGI as matching human capability across most cognitive tasks, reflecting differences in interpretation.
An AI agent describes software that can carry out multi-step tasks autonomously, such as booking services or managing workflows, often by combining multiple AI systems. Definitions vary, as infrastructure supporting these systems is still developing.
Reasoning, Compute, And Deep Learning
Chain-of-thought reasoning refers to breaking complex problems into smaller steps to improve accuracy, especially in logic or coding tasks. This method can increase response time but often produces more reliable outputs.
Compute refers to the processing power required to train and run AI models, typically provided by hardware such as GPUs, CPUs, and specialized accelerators. This computational capacity underpins the development and deployment of modern AI systems.
Deep learning is a subset of machine learning that uses layered neural networks to identify patterns in large datasets. These systems can learn features automatically and improve through repeated training, though they require large amounts of data and significant processing time.
Diffusion, Distillation, And Fine-Tuning
Diffusion models generate content by learning how to reconstruct data from noise, a process used in many image, audio, and text generation systems.
Distillation involves transferring knowledge from a larger “teacher” model to a smaller “student” model, enabling more efficient systems with similar behavior. This approach can reduce computational requirements while maintaining performance.
Fine-tuning refers to additional training applied to a pre-trained model to improve performance in specific tasks or domains, often using targeted datasets.
GANs And Hallucinations
Generative adversarial networks, or GANs, consist of two models that compete with each other: one generates data while the other evaluates it. This structure helps improve realism in outputs such as images or videos.
Hallucinations occur when AI systems generate incorrect or fabricated information. These errors are often linked to gaps in training data and remain a significant challenge, particularly for general-purpose models.
Inference, LLMs, And Memory Caching
Inference is the process of running a trained model to generate predictions or responses. It relies on prior training and can be performed on a range of hardware, from mobile devices to cloud-based systems.
Large language models, or LLMs, power many AI assistants, including systems like ChatGPT, Claude, Gemini, Copilot, and Le Chat. These models use billions of parameters to understand and generate language by predicting sequences of words.
Memory caching improves efficiency by storing previously computed results, reducing the need for repeated calculations and speeding up responses.
Neural Networks And RAM Constraints
Neural networks are the underlying structures that enable deep learning, inspired by interconnected systems in the human brain. Advances in GPU hardware have allowed these networks to scale and improve performance.
The term “inRAMageddon” refers to growing shortages of RAM chips driven by demand from AI data centers. This shortage has increased costs across industries, including gaming, consumer electronics, and enterprise computing.
Training, Tokens, And Transfer Learning
Training is the process of feeding data into a model so it can learn patterns and produce useful outputs. It is resource-intensive and often requires large datasets.
Tokens are the units of data processed by language models, representing segments of text used for both input and output. Token usage is also a key factor in how AI services are priced.
Transfer learning allows a model trained on one task to be adapted for another related task, improving efficiency when data is limited.
Weights And Model Behavior
Weights are numerical parameters within AI models that determine the importance of different inputs during training. They are adjusted over time to improve accuracy, shaping how a model interprets data and generates outputs.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.