The plain-English version
An optimization approach that updates model parameters in a direction intended to reduce a loss.
Why it matters
The loss definition, learning rate, and data handling influence what the model learns.
Let’s make it click
A model adjusts weights using a batch of labeled examples and the gradient of its error.
Vibe Bro’s heads-up
A decreasing training loss does not prove the chosen objective matches user needs or that evaluation data remains unseen.
The docs behind it
Documentation links checked .
Editorial review · October 28, 2026
Supported: Gradient descent uses loss gradients to update parameters; reducing training loss is not claimed to establish task quality.