Test-Time Compute vs Training-Time Compute

Definition

Distinction in AI systems between resources spent creating a model (training) vs. resources spent running it (inference/test-time). Recent advances allow directing more compute to inference (letting the model 'think' longer) rather than just scaling training, creating a new scaling dimension parallel to traditional training-time scaling.

Cast within

No topic-region cast yet — this appears once Test-Time Compute vs Training-Time Compute's compiled claims are aligned into a topic region's argument tree.

Claims mentioning Test-Time Compute vs Training-Time Compute (2)

Your notes

Loading notes...