A cutoff is a property of a model version, not a claim that it remembers every earlier fact. Anthropic’s model overview lists a reliable knowledge cutoff separately from a training-data cutoff; the dates can differ for a model (Anthropic, Models overview). Read the provider’s documentation for the specific model you use rather than asking the model to guess its own freshness.
For example, an assistant explaining a library release published after its model’s cutoff needs the release notes. Give it those notes or a tool that retrieves them, and require links to the relevant passages. Test a question whose answer changed between releases. The useful check is whether it uses the supplied current evidence rather than repeating an older interface from training.
Retrieval provides information at answer time. Lewis and colleagues’ RAG paper combined a pretrained model with an external document index, separating stored model parameters from retrievable evidence (Lewis et al., RAG). Supplying a current document in a prompt does not itself retrain the model. That distinction matters when a product answers both current public questions and questions about private material.
For builders, make freshness explicit: record the document version, publication date and retrieval time where the task needs them. Treat missing evidence as a reason to retrieve or qualify an answer. What is RAG? covers the retrieval pipeline; context window defines where the evidence goes. How to check an AI answer offers a practical source-checking workflow.
Sources
- Anthropic, Models overview — Primary source for the explanation and distinctions in this entry.
- Lewis et al., RAG — Primary source for the explanation and distinctions in this entry.