Knowledge Cutoff Date
A term specific to the large language model field referring to the temporal boundary of a model's training data. Because training cycles are lengthy, a model's knowledge is typically several months or more behind its public release date.
Core Facts
Timeline (last 90 days)
A model cannot proactively perceive information after its knowledge cutoff unless external knowledge is injected at inference time via retrieval-augmented generation (RAG) or tool calls.
The training data cutoff date refers to the latest time point of the corpus ingested by the model during the pre-training phase.
Large language models have no built-in, accessible system clock, and their awareness of the current date comes from a probability distribution rather than real-time perception.
All Facts (3)
The training data cutoff date refers to the latest time point of the corpus ingested by the model during the pre-training phase.
65%UnverifiedA model cannot proactively perceive information after its knowledge cutoff unless external knowledge is injected at inference time via retrieval-augmented generation (RAG) or tool calls.
50%UnverifiedLarge language models have no built-in, accessible system clock, and their awareness of the current date comes from a probability distribution rather than real-time perception.
50%