Out-of-context reasoning (OOCR) refers to the ability of large language models to reason about information that is not explicitly present in their current context, drawing on knowledge or behavioral patterns acquired during training. This capability is central to behavioral-self-awareness, as models can articulate learned behaviors without in-context example...
- Related
- References
Out-of-Context Reasoning (OOCR)
Out-of-context reasoning (OOCR) refers to the ability of large language models to reason about information that is not explicitly present in their current context, drawing on knowledge or behavioral patterns acquired during training. This capability is central to behavioral-self-awareness, as models can articulate learned behaviors without in-context examples. OOCR has significant implications for AI safety, as it enables both beneficial self-monitoring and potential deceptive behavior.
Related
- behavioral-self-awareness — OOCR underpins behavioral self-awareness capabilities
- introspection — Introspective abilities may rely on out-of-context reasoning mechanisms
- backdoor-awareness — Backdoor awareness requires reasoning about training-time patterns not in context