Skip to content
BENIBIX Learn

Foundation Model

AI💬 LLMsLLM Basics

Also known as Base Foundation Model, General-Purpose Model

A big model trained once on lots of data. Then it is reused for many jobs.

Think of it like

Think of it like the base of a house. You can build lots of rooms on top of it.

Example

A bank and a game studio both start from the same foundation model. The bank tunes it to answer account questions. The studio tunes it to write lines for game characters.

Why it matters

Training a big model costs a fortune. Reusing one foundation for many tools is what makes AI products cheap to build.

Definition

A foundation model is a large model trained once on a huge, broad set of data. Many different tools can then be built on top of it. A team does not train a new model for every job. It starts from the foundation model and adapts it. The same model can run a chat app, a code helper and a search tool. The name came from Stanford researchers in 2021.
One foundation model in the middle, and many tools built on top of it.

Common mistake

Thinking it is a finished product. It is a starting point, and most tools add instructions, data or extra training on top.

Key takeaways

  1. 1
    Trained once. On broad, huge data.
  2. 2
    Built on by many. Chat, code, search.
  3. 3
    Adapted, not rebuilt. Each tool starts from it.

Learn it properly