# What is Scaling Laws? Definition & Technical Mechanics

> **Source:** [SPIDITS AI AI Glossary](https://spidits.com/ai-glossary)  
> **Category:** Model Training  
> **Canonical Citation:** [https://spidits.com/ai-glossary/scaling-laws](https://spidits.com/ai-glossary/scaling-laws)

## Definition
Scaling Laws describe empirical mathematical power-law relationships predicting that an AI model's performance scales predictably as compute budget, training dataset size, and parameter count are scaled up.

## Architectural Mechanics & Deep Dive
Scaling laws in deep learning describe the empirical mathematical relationship between a model's performance and three core variables: model parameter size, training dataset size, and total compute budget. Pioneered by OpenAI and DeepMind, scaling laws demonstrate that cross-entropy loss decreases predictably as power-law functions of these metrics, guiding the multi-million dollar resource allocations of modern frontier model pre-training.

## Industry Applications & Real-World Use Cases
Compute budget allocation, pre-training parameter design, and benchmark estimation

## Frequently Asked Questions

### What are the Chinchilla scaling laws?
A landmark 2022 finding stating that for optimal training, model parameter size and training tokens should scale in equal proportion (making many historical models under-trained on text).

### Do scaling laws have a limit?
There is active debate; researchers are hitting resource ceilings regarding available high-quality human text and physical power grids.

## Related Terminology
- [parameters](https://spidits.com/ai-glossary/parameters)
- [deep-learning](https://spidits.com/ai-glossary/deep-learning)
- [gpu](https://spidits.com/ai-glossary/gpu)

---
*From the SPIDITS AI Knowledge Base (336+ terms indexed). Explore full technical definitions: [https://spidits.com/ai-glossary](https://spidits.com/ai-glossary)*