capabilities

Definition

The set of things a model can actually do, usually carved into measurable bundles: reasoning, coding, tool use, long-context retrieval. Much of the field's empirical work is about how these scale with parameters, data, and post-training. The word also carries a safety sense — dangerous capability evaluations probe for skills a model should not have — which is why labs report capability profiles rather than a single score.

Explainers using this term

• Paper walkthrough: CyberFactory — turning wild CVEs into runnable training problemsCyberFactory: Scaling Cyber Security Capabilities with Instances from the Wild
• Paper Walkthrough: GigaBrain-0.7 — 37,000 Hours of Embodied Experience and a Three-System Robot BrainGigaBrain-0.7: Scaling Embodied Foundation Models to Emergent Capabilities with a Three-System Architecture