Embedded Files
AI Hallucination
Image created by Martin Ågren using the AI-model "DALL-E".
BRIEFLY About AI Hallucinations
What Is an AI Hallucination?
🔹When AI gives you incorrect information = AI “hallucinates.”
Cause
🔹AI has not been trained sufficiently on the subject and
🔹Does not have access to the necessary information.
Prevent
🔹Online search = Tip: You can also iterate: “Search again online”
🔹Paste in text
🔹File: Upload
Check Link
🔹Does the link work? Hover over the link: Hand pointer? Clickable?
🔹Does the link lead to a page that exists?
🔹Does the link lead to a page that is relevant?
Check Quotes
🔹Ask AI to support its claim with quotes from the text.
🔹Check that the quote exists in the text.
Written by Martin Ågren - March 18, 2025
WHAT Are AI Hallucinations?
Simple explanation: When AI generates incorrect information. This is called an "AI hallucination".
Technical explanation: GenAI model predicts tokens in a way that makes the output contain incorrect information.
CAUSE of AI Hallucinations?
Insufficient pre-training and a lack of sufficient, relevant information. In other words, if an AI chatbot has not been exposed to enough material (or information through reference text or documents or web search) on a particular subject, it may generate an incorrect answer.
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.
Internet search – Press Search button or Instruct AI to search online: “Search multiple sites online and provide clickable source link.”
Iterate – For example: "Search online again." This can help the AI eventually find the necessary information.
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.
Short conversations – Avoid excessive details; keep input and output concise.
Temperature setting (for advanced users) – Lower temperature reduces overly creative responses. Useful if you use AI via an API or run local inference.
IDENTIFYING AI Hallucinations
Check links – Hover over links; if they’re not clickable, the information may be unreliable. If the page exist: is it relevant?
Check source info – Click source links and read content to confirm accuracy.
AI self-checking – Ask AI to reflect on its response for errors.
Citations – Request quotes from source: “Make a citation by quoting word for word, from your source…”. Then check if that quote actually exists in the source text.
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.
For MANAGERS and LEADERS
As a manager or leader, you should ensure that AI-generated content is never used in your organization's operations without appropriate review. Everyone in the organization should approach AI-generated content with a critical mindset, but as a leader, you have a greater responsibility to reduce the risks associated with AI hallucinations.
Establish simple and clear procedures that define when, how, and by whom AI-generated content should be reviewed, and consider documenting these procedures in your AI policy.
Lead by example by demonstrating how AI-generated responses are verified and questioned in your own work. This helps set the standard and encourages others in the organization to adopt the same constructive and preventive approach. Make critical review a natural part of your workflow so your organization can use AI safely and responsibly.
Page updated
Google Sites
Report abuse