AI Glossary

AI Glossary: Industry Concepts

Fearghal

08 May 2026 — 1 min read

Understanding the AI landscape helps designers navigate vendor decisions, cost considerations, and market dynamics. Here are a bunch of foundational concepts that we should all have some awareness of. No need to learn them by heart, but this is a good reference manual with references where possible.

Foundation Models

Large-scale models trained on vast, diverse data serving as bases for many downstream applications. Stanford researchers introduced the term in 2021. Building requires massive resources; using them is far cheaper—democratizing AI access.

Reference: Bommasani, R., Hudson, D.A., Adeli, E., Altman, R. et al. (114 authors), "On the Opportunities and Risks of Foundation Models", Stanford CRFM, 2021

Frontier Models

The most advanced models pushing capability boundaries with potentially significant risks. The Frontier Model Forum defines these as exceeding "capabilities present in today's most advanced existing models." Subject to increased regulatory scrutiny.

Reference: Anthropic, Google, Microsoft, and OpenAI, "Frontier Model Forum Announcement", July 2023

Open-Source vs. Closed Models

Distinction between publicly available weights (open: Llama, Mistral) versus API-only access (closed: GPT-4, Claude). Open enables customization and self-hosting; closed offers convenience and safety features. Many organizations adopt hybrid approaches.

Reference: ⚠️ No single authoritative academic paper — distinction emerged organically with releases like Meta's LLaMA (2023).

Inference Costs / Tokens-Per-Dollar

Economic metrics for production AI—typically per million tokens processed, ranging from $0.10 to $150+ per million output tokens. Key metric for high-volume applications. Optimization: model selection, caching, quantization, batching.

Reference: Epoch AI, "Inference economics of language models", 2024
Additional: Andreessen Horowitz, "LLMflation - LLM inference cost is going down fast", 2024

Fine-Tuning Services

Platforms enabling model customization without building ML infrastructure—OpenAI, Vertex AI, AWS Bedrock. Appropriate when you need consistent behavior or domain expertise that prompting can't achieve. Costs range from hundreds to thousands of dollars.

Reference: Hadi et al., "FineTuneBench: How well do commercial fine-tuning APIs infuse knowledge into LLMs?", 2024

This glossary is part of a series covering AI and LLM concepts for product designers.

AI Glossary: Infrastructure & Deployment

Understanding infrastructure helps designers appreciate performance constraints, cost considerations, and deployment options that affect product decisions. Hardware & Serving GPU (Graphics Processing Unit) Specialized processor for parallel calculations, dominant hardware for AI. GPUs excel at thousands of simultaneous calculations essential for neural networks. NVIDIA's H100 and A100 are

AI Glossary: Human-AI Interaction & UX Design

This is probably the most important part of the glossary for designers and these patterns and principles are directly relevant to product design practice. Understanding them enables the creation of AI experiences that are useful, usable, and appropriately trusted. Interaction Patterns Conversational UI Interface paradigm where interaction occurs through natural

AI Glossary: Safety, Alignment & Evaluation

These concepts are critical for responsible AI product development—understanding failure modes, safety measures, and evaluation approaches directly informs design decisions. Safety Concepts AI Safety The interdisciplinary field preventing accidents, misuse, or harmful AI consequences. Encompasses alignment research, risk monitoring, robustness, and developing norms for beneficial AI operation. Reference: Amodei,

AI Glossary: Training, Fine-Tuning & Data

Understanding how models are built and adapted helps designers recognize capability sources, appreciate customization options, and understand training-related behaviors and limitations. Pre-Training Concepts Pre-training The initial, computationally intensive phase of training on massive unlabeled text. During pre-training, models learn language patterns, grammar, factual knowledge, and reasoning capabilities by processing billions