Model Weights
Model weights are the learned numbers that set the strength of connections in a neural network and together encode everything the model has learned.
What is Model Weights?
Whoever holds a model's weights can run, copy or modify it, so frontier developers treat weights as highly sensitive assets. Policy debates about open-weight release and security controls centre on who should be able to access them.
How it works
Weights are stored as large files of numbers. Loading them into compatible software reproduces the trained model exactly.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions