DalšíDalší průvodce
BloombergGPT a finanční LLMs
Jazyk AI
Jazyk AI GUIDE
A knowledge cutoff is the approximate date after which a language model's training data contains little or no information, so the model has no built-in knowledge of later events.
Poslechněte si tento návodAI-vyprávěný · ~3 min
It matters because a model may confidently present outdated prices, software versions, officeholders or research as current. That happens unless it is connected to search or given fresh documents.
Large language models learn from a snapshot of text collected up to some point. They then go through further training and safety work before release. The knowledge cutoff marks the end of that snapshot. Release often comes months after the cutoff, and a model stays in use long after release, so the gap between what it knows and the present keeps widening. The cutoff is fuzzy rather than a hard line. People keep writing about events long after they happen, so the final months before a cutoff are thinly represented in the training data. This is one reason models are often unsure of their own cutoff, or underestimate it. Vendors usually publish the date in model documentation. Asking the model itself is not a reliable way to find it. Stale knowledge shows up in predictable ways: - outdated version numbers and API syntax - old prices - former officeholders or executives - superseded guidelines - no awareness of newer models, including the model's own successors A model may also assume the current date is close to its cutoff. That distorts its reasoning about ages, deadlines and what counts as recent. The main workarounds supply fresh information when the question is asked. Web search tools let the model read current pages. Retrieval-augmented generation pulls from a document store that someone keeps up to date. Simply stating the current date in the system prompt fixes many date errors. Fine-tuning can add knowledge, but it is slow and costly compared with retrieval as a way to keep facts current. A common misconception is that a chatbot with browsing has a later cutoff. Its training cutoff is unchanged. It is reading new sources, and its answers are only as current and accurate as the pages it retrieves.
Someone asks a model with no browsing for the latest version of a software library. It names the version that was current when its training data was collected.
A model asked what year it is guesses a year near its training data unless the app puts today's date in the system prompt.
A news assistant runs a web search, reads articles published this week and cites them, which lets it answer about events after its cutoff.
A company's internal assistant uses retrieval over policy documents that are re-indexed every night. Its answers reflect current policy whatever the model's cutoff.
Model releases have become more frequent, which narrows the typical gap between cutoff and use, but no trained model can be fully current. The more lasting trend is tighter integration of search and retrieval, so current facts come from sources the user can check. Some research explores editing or updating knowledge inside a model without full retraining, though these methods have known limits. For users, the practical habits stay the same. Check the stated cutoff, notice when a question depends on recent events, and verify time-sensitive answers against dated sources.
Newsletter zdarma
Tři ověřené příběhy ze světa AI každý všední den ráno, psané srozumitelnou angličtinou. Navždy zdarma, bez reklam.
Jeden e-mail každý všední den. Odhlásit se můžete jedním kliknutím. Vaši adresu nikdy neprodáváme ani nesdílíme.
Otestujte se
Okamžitá zpětná vazba ke každé odpovědi a sdílitelný certifikát s ověřitelným ID po absolvování kurzu.
Podpořte bezplatnou AI vzdělávání. AI Understanding je nezisková organizace 501(c)(3) — žádné reklamy, žádná platba za přístup, nikdy. Darujte
A knowledge cutoff is the approximate date after which a language model's training data contains little or no information, so the model has no built-in knowledge of later events. It matters because a model may confidently present outdated prices, software versions, officeholders or research as current. That happens unless it is connected to search or given fresh documents.
The cutoff marks the end of the training-data snapshot. Release usually comes later, and the model is used long after that.
Coverage of any period keeps growing for years. At training time the most recent months have relatively few documents, so the model sees little about them.
Vendors usually publish the cutoff. The model's own answer is unreliable for the reasons covered in the guide.
Aniž by mu to bylo sděleno, model má tendenci předpokládat, že datum je blízko jeho meze. Uvedení dnešního data to přímo potvrzuje.
Procházení přidává nové zdroje v okamžiku otázky. Nemění to, co se model naučil při školení, a načtené stránky mohou být samy o sobě špatné.
Učte se dál
Pro toto téma bylo vybráno více průvodců
DalšíDalší průvodce
BloombergGPT a finanční LLMs
Jazyk AI