Technical
Parameter
The learnable weights and biases in a neural network that are adjusted during training. Model size is often expressed in parameter count (e.g., 7B, 70B, 405B).
Explained at five levels
Level 1
A tiny knob inside the AI's brain that gets adjusted during learning — billions of these knobs working together make the AI smart.
Level 2
The numbers inside a neural network that get tweaked during training. More parameters generally means a more capable model — GPT-4 has hundreds of billions.
Level 3
The learnable weights and biases in a neural network that are adjusted during training. Model size is often expressed in parameter count (e.g., 7B, 70B, 405B).
Level 4
The trainable scalar values (weights and biases) in a neural network, collectively defining the function the model computes. Parameter count is a primary scaling dimension correlated with model capability.
Level 5
The elements of the parameter vector θ ∈ ℝⁿ defining the model's learned function — with scaling laws establishing power-law relationships between parameter count, training compute, dataset size, and loss.
Definitions are educational summaries. Terminology can vary by source and context.
Sources
- NIST Trustworthy and Responsible AI Resource Center: Glossary — National Institute of Standards and Technology. Accessed 2026-07-20.