When you ask Leo certain types of questions you will expect the response to apply knowledge of current events, but Leo may require information it doesn’t have available from its initial training in order to answer the question.

By searching for information and using the results in its response, Leo can improve the accuracy and relevance of its responses, especially for queries where the language model might otherwise be prone to generating unreliable information, or even “hallucinations”.

How does it work?

Leo uses the Brave Search API to collect relevant information. Leo then adds the search results to the user’s prompt and sends them together to Leo's language model. Leo will then have relevant current information to consider when formulating a response to the prompt.

You can learn more about how the models work and what steps Brave takes to protect your privacy in this doc: What are the differences between Leo's AI Models?

Of course, as explained when users opt-in to use Leo, the accuracy of responses is not guaranteed, and may include inaccurate, misleading, or false information. It’s always wise to check sources directly to confirm the information you are seeking.