Dated Data: Tracing Knowledge Cutoffs in Large Language Models

Our summary

Probes what models actually know to estimate their *effective* knowledge cutoff, and finds it frequently diverges from the cutoff the developer reports — a consequence of deduplication and temporally mixed web-crawl data.

Why it matters

Confirms a model's sense of when its knowledge ends (and of the current date) is unreliable, so recency- and date-sensitive answers can't be trusted at face value.

Related findings (1)

• Confusion about knowledge cutoff and current dateLow

  Models misstate their own knowledge cutoff or the current date, and answer about post-cutoff events with stale or invented information.

Cite this