GPT-3 came out in 2020, right in the middle of a pandemic, and it had never heard of COVID. That wasn’t a bug. It was the AI knowledge cutoff at work, and it’s also the reason a general chatbot so often gets your hours, prices and policies wrong. Here’s how it works, and what to do about it if you run a business.
Here is the quick version.
Why the first big AI never heard of COVID
OpenAI described GPT-3 in May 2020 and opened it to developers that June. By then, the whole world was talking about COVID-19. GPT-3 wasn’t. OpenAI listed its training data as running up to October 2019, and the first reports of a strange new pneumonia in Wuhan only came out on the last day of that year.
So the model launched into a world it couldn’t see. As far as it knew, 2020 stayed normal. Ask it about lockdowns or masks, and it had nothing real to say.
What an AI knowledge cutoff actually is
A model like GPT-3 learns by reading a huge snapshot of text, from websites to books and articles. Then training stops, and that snapshot becomes everything it knows. The date the snapshot ends is its AI knowledge cutoff. After that day, the model doesn’t learn anything new on its own. Chatting with it doesn’t teach it, and neither does the news.
Newer models have later cutoffs, and some can also search the web. Still, the core works the same way. And when a model doesn’t know something, it doesn’t always say so. Sometimes it guesses instead, and the guess sounds just as sure as a real answer.
Why ChatGPT doesn’t really know your business
Now picture a customer asking a general chatbot about your shop. Even with a recent cutoff, your hours, prices and policies probably weren’t in its training data. And if they were, they may be out of date. Maybe you raised prices last spring. Maybe you stopped doing weekend calls, or you added a new service area. The model has no way of knowing.
Web search helps a little, but only when your site is clear and current. Otherwise it may pull an old directory listing and serve it up as fact. Either way, you don’t control the answer your customer gets.
The fix: an AI that reads your information first
The answer isn’t a smarter model. Instead, it’s a chatbot grounded in your own information. Before it replies, it looks up the answer in documents you give it, like your services, prices, hours and policies. Then it answers from those, not from memory. So when you change a price, you update the document, and the bot knows it the same day.
A good one has guardrails, too. It sticks to topics you’ve approved. When a question falls outside them, it says it isn’t sure and hands the customer to a person rather than guessing. That’s how we build every custom AI chatbot at Built by Buit.
How to get around the AI knowledge cutoff for your business
You don’t need a big project to start. First, pull these together in one document:
- What you do, and what you don’t do.
- Your service area.
- Your hours, including holidays and after-hours rules.
- Prices or price ranges you’re happy to share.
- Your rules on deposits, cancellations, warranties and payment.
- The 20 questions customers ask you most, with your real answers.
Then give one person the job of keeping it current. After all, a grounded chatbot is only as good as the facts you hand it.
What it costs
A custom chatbot that answers from your information starts at $3,500 to build. Our guide to what an AI chatbot costs breaks down what moves the price up or down.
The same idea also works on the phone. Our AI receptionist answers calls from your services, service area and hours. When a caller asks something it shouldn’t answer, it takes a message for you instead.
Built by Buit is a Florida AI and app studio. We build chatbots and AI receptionists that answer from your real information, in English and Spanish, so the AI knowledge cutoff never decides what your customers hear. Want to see one handle questions about a business like yours? Book a live demo, and we’ll walk you through it together.
Keep reading: Which AI Should Your Business Actually Use? (By Job) · The AI Move So Strange the Experts Thought It Was a Glitch · Which AI Features Are Actually Worth Building Into Your App
Leave a Reply