Hamel HusainML Engineer Helping Companies Build AI Products

More Context Doesn't Mean Better Answers. It Means More Hallucinations.

4 Feb 2026 · 0 min

Watch on YouTube

0:24

Hamel Husain explains why dumping a year of documents into an AI model to answer a question about last week's changes backfires. Too much unfocused context confuses the model and triggers more hallucinations. The fix is targeted retrieval that matches the query's actual scope.

Giving your AI hundreds of documents for context about a user query is often a bad idea. For example, if user is asking, how did this project change in the last week? It would be a bad idea to give it documents about the last year. What you wanna do is think about retrieval carefully. If they're only asking about the last week, filter the documents. Otherwise, it's gonna be very confusing for the model and you're gonna increase the chances of hallucination.