Foundation model
Foundation model is a general-purpose AI model that is learned once and used for various purposes. It becomes the foundation on which various services are built.
The foundation model is a general-purpose AI model that is learned on a large scale from a large amount of data and used for various purposes such as translation, summarization, consultation, and coding. As the name suggests, it is the basic construction of a building, and it is a structure where various services are built on top of it through fine tuning or prompts.
In the past, translation AI and summary AI had to be created separately, but the industry has changed as it has been confirmed that one large model can perform multiple tasks. LLMs such as GPT and Claude are representative examples, and an ecosystem has been formed where numerous startups build services on the foundation model created by a few companies.
As the foundation is concentrated in a few companies, concerns about dependency and monopoly are also raised. Another risk factor in this structure is that if there is a flaw or bias in the underlying model, it will spread to all services above it.
✅ Why it matters
- Its strength is its versatility in responding to a variety of tasks with one model
- The ability to quickly build services on top of it has fostered an AI startup ecosystem
- It is a key concept in understanding the hierarchical structure of the current AI industry.
⚠️ Limits and debates
- A huge amount of capital is invested in development, concentrating power in a few big techs
- Defects and biases in the basic model are propagated to all higher-level services
- A general-purpose model may be worse than a specialized model in a specific area of expertise