A capability is an ability or functional competence that a system can demonstrate. For LLMs, capability describes what the model can produce or accomplish, while reliability asks how consistently and accurately it does so. section 1.9 section 7.0
LLM capabilities span several kinds of work:
- Translate languages, transcribe speech, read text in images or printed documents, and support search or information retrieval. section 9.1
- Generate new language and, in broader generative AI applications, images, music, or videos. section 9.1 section 1.9
- Handle many tasks through one model, although broad capability does not guarantee dependable results or human-like understanding. section 9.1 section 7.0
Capability versus certainty
An LLM may translate text, summarize information, or generate an answer, yet fluent output is not by itself evidence of intelligence, factual accuracy, or reliable reasoning. section 7.0 section 1.9
