Scaling Laws
Scaling laws are observed relationships showing that AI model performance improves predictably as model size, training data and computing power increase.
What is Scaling Laws?
Research from OpenAI in 2020 and DeepMind's 2022 Chinchilla study showed smooth gains from scale and how best to balance parameters and data. Scaling laws underpin the large investments in AI compute and the use of compute thresholds in regulation.
How it works
Performance, measured by loss, follows a power-law curve as compute, parameters and data increase.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions