What is a large language model (LLM)?
A large language model (LLM) is an AI system trained on vast amounts of text to understand and generate human language, able to answer questions, summarise, translate and write. It predicts likely text based on patterns it learned, which makes it fluent but also prone to inventing facts if ungrounded.
A large language model learns statistical patterns of language from enormous text corpora, so it can interpret a request and generate a fluent response. LLMs power chatbots, writing assistants and question-answering tools.
Their strength is language understanding and generation; their weakness for data work is that, on their own, they generate plausible text rather than retrieve verified facts, which is why an ungrounded LLM can hallucinate numbers that were never in your data.
iDBQuery uses a language model as one component of a carefully grounded system, and deliberately keeps the specific model behind the scenes. The model's job is to understand your plain-language question and help translate it, but the answer does not come from the model's imagination. iDBQuery converts your question into SQL that runs against your live data, so figures are computed from your real records, and then cites every result back to the exact source rows. That architecture, language model for understanding, real queries plus citations for facts, is what makes iDBQuery trustworthy where a bare LLM chatbot is not. You get the natural-language ease of an LLM without the risk of made-up numbers.
Updated 2026-06-22