raw

Definition

Input taken as it comes — pixels, waveforms, bytes, unedited text — with no hand-designed feature extraction in front of it. Claiming to learn from raw input is claiming that the model discovers the representation itself, which is the central selling point of deep learning over pipelines built on expert-designed features. The trade-off is that raw input needs far more data and compute to make the same signal usable.

Explainers using this term

• Speech Synthesis from Scratch — From Text to a VoiceWaveNet: A Generative Model for Raw Audio
• Paper walkthrough: StateM — 95.3% on Terminal-Bench 2.1 and a USD 15 run, without touching a single weightStateM: Reaching 95.3% Raw Accuracy, or a \$15 Frontier Run, on Terminal-Bench 2.1 via Harness Scaling