AI Hallucination
Incorrect output from AI

Written by Martin Ågren - March 18, 2025

What are AI Hallucinations?

• Simple explanation: When AI generates incorrect information. 

• Technical explanation: GenAI model predicts tokens in a way that makes the output contain incorrect information. 

What is Causing it?

Insufficient pre-training or lack of available information. 

Prevention Methods

• English – Usually safest to write in English, except when the topic is primarily discussed in Swedish.  

• Open-ended questions – Prevent AI from being misled or fixated on specific details.  

• Short conversations – Avoid excessive details; keep input and output concise.  

• Internet search – Press Search button or Instruct AI to search online: “Search multiple sites online and provide clickable source link.”  

• Paste reference text – Provide relevant text at the start of the chat and instruct AI to use it.  

• Upload files – Upload relevant documents for AI to reference.  

• Temperature setting – Lower temperature reduces overly creative responses. 

Fact-Checking & Identifying AI Hallucinations

• Check source links – Hover over links; if they’re not clickable, the information may be unreliable.  

• AI self-checking – Ask AI to reflect on its response for errors.  

• Direct citations – Request exact word-for-word quotes from sources: “Make a citation by quoting word for word, from your source…”  

• Verify sources – Click source links and read content to confirm accuracy.  

• Experience – The more you use AI, the easier it is to detect hallucinations.  

• Bias awareness – Different AI chatbots may provide different answers due to bias.  

• Compare AI responses – Cross-check answers from multiple AI chatbots.  

• Your expertise – Errors are easier to spot in topics where you have strong knowledge.