What is the core idea behind foundation models?
A foundation model is a general-purpose base you adapt, not a single-purpose tool you build from scratch.
How do foundation models differ from related concepts?
| Concept | Difference |
|---|---|
| Foundation Model vs LLM | Large language models are foundation models for text. Foundation models also include image, audio, and multimodal models. |
| Foundation Model vs Fine-Tuned Model | A foundation model is the general base. A fine-tuned model is that base adapted to a specific task. |
| Foundation Model vs Narrow AI | Narrow AI is built for one task. A foundation model is built to be adapted to many. |
How do foundation models work?
- The model is pre-trained on vast, broad datasets
- It learns general-purpose representations of language, images, or other data
- Developers adapt it through prompting, fine-tuning, or retrieval
- One base model can power many different applications
What are examples of foundation models?
- GPT, Claude, and Gemini for language
- Stable Diffusion and DALL-E for images
- Multimodal models that combine text, image, and audio
- Open-weight models such as Llama and DeepSeek
Why are foundation models important?
Foundation models shifted AI from building a new model per task to adapting one powerful base model to many. This is the economic and technical engine behind the current AI boom. The term was coined by Stanford researchers in 2021.
How are foundation models used in practice?
Foundation models underpin chatbots, coding assistants, image generators, and enterprise AI platforms. Concerns include concentration of power among a few labs, cost, bias, and the difficulty of auditing what they have learned.
Frequently Asked Questions
Is a foundation model the same as a large language model?
Not exactly. All large language models are foundation models, but foundation models also include image, audio, and multimodal systems, not just text.
Who makes foundation models?
Mainly large AI labs such as OpenAI, Anthropic, Google, Meta, and DeepSeek, because training them requires enormous data and compute.