Back to glossary
AI
Hallucination
When an LLM generates information that isn't real — made-up prices, stock, customer names.
Hallucination is the biggest LLM weakness. LeadsUp uses a post-processing hallucination guard: every number in a bot reply (price, sqm, stock) is auto-validated against your DB. If a non-matching number appears, the message is blocked and the conversation is transferred to a human.
Related terms
LLM (Large Language Model)
AI model trained on massive text corpora; can converse in natural language.
ReadTool use
LLM's ability to call functions/APIs in addition to generating text.
ReadRAG (Retrieval Augmented Generation)
LLM retrieves relevant documents from a knowledge base before generating its reply.
ReadWant to see how it works in practice?
14 days free. See these concepts live on your own bot.