What is a parameter?

Parameters are the internal settings that AI models use to make predictions. They're like the knobs and dials that control how the model behaves.

Parameters are numbers stored in the model that determine how it processes information. During training, these numbers are adjusted so the model learns patterns from data.

[Think of it like]: The weights in a neural network that determine how strongly different connections influence the output.

Models are often described by their parameter count:

• ▸[Small models]: Millions of parameters (7B = 7 billion)
• ▸[Medium models]: Tens of billions (70B = 70 billion)
• ▸[Large models]: Hundreds of billions (175B = 175 billion parameters in GPT-3)

[More parameters] generally means:

• ▸More capable models
• ▸Better performance on complex tasks
• ▸Slower inference
• ▸Higher costs
• ▸More memory required

[During training]: Parameters are adjusted to minimize errors [During inference]: Parameters are fixed—they determine how the model responds

The model uses these parameters to:

• ▸Understand input
• ▸Process information through layers
• ▸Generate output

[Capability]: More parameters can store more knowledge and patterns [Cost]: Larger models cost more to train and run [Speed]: Larger models are slower [Requirements]: Need more powerful hardware

[GPT-3.5]: ~175 billion parameters [GPT-4]: Estimated in the trillions (exact number not disclosed) [Claude 3]: Estimated hundreds of billions [Smaller models]: 7B, 13B, 70B parameters

[Parameters]: Internal model settings (fixed after training) [Tokens]: Units of text processed (varies per request)

Don't confuse these! Parameters are about the model's size, tokens are about how much text you're processing.

Most users don't need to worry about parameters directly. What matters is:

• ▸[Model performance]: Does it do what you need?
• ▸[Speed]: Is it fast enough?
• ▸[Cost]: Is it affordable?
• ▸[Quality]: Are the results good?

When choosing a model:

• ▸[Larger models]: Better for complex tasks, more expensive
• ▸[Smaller models]: Faster, cheaper, good for simple tasks
• ▸[Right-sized]: Choose based on your needs, not just size

Parameters are the technical foundation of how AI models work, but for most practical purposes, you care more about what the model can do than how many parameters it has.