Foundation Models
Foundation Models are large-scale AI models trained on broad, diverse datasets using self-supervised learning, which can then be adapted to a wide range of downstream tasks. The term was coined by Stanford HAI's Center for Research on Foundation Models (CRFM) in 2021 to describe models like GPT, CLIP, and DALL-E that serve as a general-purpose starting point for many applications. They unify text, vision, audio, and multimodal tasks under a single pretrained backbone that is fine-tuned or prompted for specific use cases.
Every major AI product in 2026 — from enterprise copilots to autonomous agents — is built on top of foundation models, making fluency with them a baseline expectation for AI engineers and applied researchers. Companies hiring in this space look for engineers who can evaluate, adapt, serve, and reason about the tradeoffs of different foundation models at production scale. Understanding foundation models is also essential for responsible AI work, as their training data, biases, and emergent capabilities directly determine downstream product risks.
🎓 Courses
Generative AI: Foundation Models and Platforms
by IBM Skills Network
Directly covers the concept of foundation models from first principles, including LLMs, GANs, VAEs, transformers, and diffusion models. Hands-on labs use IBM watsonx and Hugging Face, making it practical from day one.
Open Source Models with Hugging Face
by Hugging Face team
Teaches you to find, load, and run open-source foundation models from the Hugging Face Hub for text, audio, image, and multimodal tasks — exactly the workflow used by practitioners building on top of foundation models.
CS324: Large Language Models (Lecture Notes)
by Percy Liang, Tatsunori Hashimoto, Christopher Ré
The canonical academic course on foundation models from the team that coined the term. Covers modeling, theory, ethics, and systems aspects at a rigorous level. Lecture notes are freely available and remain highly relevant.
CS336: Language Models from Scratch
by Percy Liang et al.
Walks through the full pipeline of building a language model from data collection and preprocessing through transformer implementation, training, and evaluation — the deepest public course on how foundation models actually work internally.
Open Source Models with Hugging Face (Short Course on Coursera)
by Hugging Face team
A concise, hands-on guided project for those who want a quick, structured introduction to running foundation models via the Transformers library without a lengthy time commitment.
📖 Books
AI Engineering: Building Applications with Foundation Models
Chip Huyen · 2025
The most read book on the O'Reilly platform in 2025. Covers the full lifecycle of building production systems on foundation models: prompt engineering, RAG, fine-tuning, agents, evaluation, latency, and cost optimization. Written for engineers, not researchers.
Introduction to Foundation Models
Pin-Yu Chen, Sijia Liu · 2025
A rigorous academic treatment published by Springer covering trustworthiness, jailbreak attacks and defenses, watermarking, backdoor risks in diffusion models, and red-teaming — essential for anyone building safe and robust systems with foundation models.
🛠️ Tutorials & Guides
The Full Guide to Foundation Models
A well-structured, beginner-friendly guide covering what foundation models are, their architecture types (LLMs, GANs, VAEs, multimodal, vision), and real-world use cases. Good conceptual foundation before diving into code.
Reflections on Foundation Models
The original Stanford HAI piece that defined the term 'foundation model' and set the research agenda for the field. Essential reading for understanding the historical context, capabilities, and societal risks articulated at the field's founding.
Foundation Models 101: Guide & Essential FAQs
Answers the most common practitioner questions about foundation models — what makes them different from traditional ML models, how fine-tuning and adaptation work, and when to use them versus training from scratch.
Learning resources last updated: June 18, 2026