A · BROAD CATEGORY
Foundation Models
Broadly trained reusable models.
Large Language Models
Language-focused foundation models.
03 · Foundation Models & LLMs
Foundation Models: One Base Idea, Many Capabilities
Foundation models are broadly trained models that can be reused across many tasks, including language, vision, audio, multimodal, and scientific applications.
Understand the difference between a broad model family, its specializations, and the learned parameters inside a model.
The Big Picture · model family map
FOUNDATION MODELSBroadly trained and reusable
LLMLanguage-focused foundation models
VISIONBroad visual representations
AUDIOSpeech and sound
MULTIMODALMultiple modalities
SCIENTIFICScientific-domain data
INSIDE THE MODEL · PARAMETERS
Learned numerical values adjusted during training. Parameters are inside models, not another model family.
Learned numerical values adjusted during training. Parameters are inside models, not another model family.
Knowledge Structure
Category, specializations, and internal property
B · SPECIALIZATIONS
Vision Foundation Models
Visual representations for visual tasks.
Speech / Audio Models
Speech and sound.
Multimodal Models
Multiple information types.
Scientific Models
Scientific-domain data.
C · INSIDE
Parameters
Learned values inside a model.
Key Distinctions
Foundation ModelBroad reusable category≠LLMLanguage-focused family
ParameterLearned internal value≠TokenInput or output unit
Multimodal ModelMultiple modalities≠LLMPrimarily language-focused
Recommended Learning Path
This is a learning order, not a technical pipeline.
01Foundation Models
Broad category.
02LLMs
One specialization.
03Specialized Families
Vision, audio, multimodal, scientific.
04Parameters
What training changes inside models.
Remember the structure
Foundation models are broadly trained reusable models. LLMs focus on language, other families specialize in modalities or domains, and parameters are learned values inside models.