Why do LLMs hallucinate? Our co-founder and CTO Simon explains, and shows how we fixed it at v7.

Added: 2026-04-29

[00:00:01]Large language models are just a statistical model trained to predict the next token from a series of tokens. They use a reflection of the training data.

[00:00:10]However, sometimes they do mistakes.

[00:00:11]That can be because the training data included errors. It can be that you didn't give enough context in your prompt for what you're trying to do. Or it can be that there wasn't a way for the model to say I don't know in the system that you created. At the DC go, we are forcing the model to point directly on the source data for any output it creates. So we can easily validate that the output is valid and the model hasn't hallucinated.