I've been using Google Notebook LM for several months and find it a highly effective digital notebook. Key features include:

  • Storage for various content types (PDFs, Google Docs, videos, websites)

    • Storage for various content types (PDFs, Google Docs, videos, websites)

  • Capacity for up to 50 items

    • Capacity for up to 50 items

  • Creation of customized podcasts, FAQs, study guides, tables of contents, timelines, and briefing documents

    • Creation of customized podcasts, FAQs, study guides, tables of contents, timelines, and briefing documents

  • Website: https://notebooklm.google/

    • Website: https://notebooklm.google/

    Personal Experience with AI Hallucinations:

    Here is an example of my Alex Hormozi Notebook LM. I am a big fan of his work and study his videos in detail. I want to create quizzes and work on retaining all his key lessons. The product does a great job of breaking down information from the 16 videos I uploaded. (Some of the videos were 4 hours long!). When asking questions about it, I did get some surprising results.  I asked the AI, “What is the CLOSER Framework”. It made up the answer!

    Here is Alex's actual CLOSER FRAMEWORK.

    I know the CLOSER FRAMEWORK was discussed in several of the videos since I had already watched them. I tried several different keywords and searches. I could not make it pull up the correct information from the data even though I knew it was there.

    Thinking this was an isolated incident, I tried again with Management Diamond and got the same issues. I tried several times with many different prompts. It came close when I typed in “Diamond” and finally got the correct information from the LM Notebook.

    Understanding AI Hallucinations:

    These experiences highlight a crucial aspect of current AI tools: hallucinations. AI hallucinations occur when models generate incorrect or misleading information while appearing confident. These errors can range from minor inaccuracies to wholly fabricated information.

    Techniques to Prevent/Reduce AI Hallucinations:

  • Provide Clear, Narrow Scope Strategy: Define specific boundaries for the task. Example: "Summarize the impact of inflation on multifamily rent pricing in 2024. Do not discuss unrelated markets."

    • Provide Clear, Narrow Scope Strategy: Define specific boundaries for the task. Example: "Summarize the impact of inflation on multifamily rent pricing in 2024. Do not discuss unrelated markets."

  • Request Sources or Citations Strategy : Ask the LLM to base its response on verifiable references. Example: "Explain the advantages of high-dividend ETFs. Cite sources or mention if verification is needed."

    • Request Sources or Citations Strategy : Ask the LLM to base its response on verifiable references. Example: "Explain the advantages of high-dividend ETFs. Cite sources or mention if verification is needed."

  • Use Structured Output Formats Strategy: Request answers in specific formats to guide the model's logic. Example: "Describe three key reasons why CAP rates are increasing. Use bullet points for each reason."

    • Use Structured Output Formats Strategy: Request answers in specific formats to guide the model's logic. Example: "Describe three key reasons why CAP rates are increasing. Use bullet points for each reason."

  • Set Limitations on the Response Strategy: Direct the model to avoid speculation. Example: "Provide a fact-based description of NOI calculation. Do not offer hypothetical scenarios."

    • Set Limitations on the Response Strategy: Direct the model to avoid speculation. Example: "Provide a fact-based description of NOI calculation. Do not offer hypothetical scenarios."

  • Use Multi-Step Prompts to Verify Logic Strategy: Ask the model to walk through processes step-by-step. Example: "Explain step-by-step how to calculate net rental yield for a property."

    • Use Multi-Step Prompts to Verify Logic Strategy: Ask the model to walk through processes step-by-step. Example: "Explain step-by-step how to calculate net rental yield for a property."

  • Prompt for Confidence Levels Strategy: R equest the model to include confidence levels with its information. Example: "What are the latest trends in multifamily financing? Provide answers with confidence levels (High, Medium, Low)."

    • Prompt for Confidence Levels Strategy: R equest the model to include confidence levels with its information. Example: "What are the latest trends in multifamily financing? Provide answers with confidence levels (High, Medium, Low)."

  • Limit the Model to Verified Knowledge Strategy: Instruct the model to rely only on specific sources or knowledge. Example: "Explain the impact of rising interest rates on apartment valuations, based only on data available up to 2024."

    • Limit the Model to Verified Knowledge Strategy: Instruct the model to rely only on specific sources or knowledge. Example: "Explain the impact of rising interest rates on apartment valuations, based only on data available up to 2024."

  • Provide Real-World Constraints Strategy: Give contextual boundaries to focus the response. Example: "Given that inflation is over 3%, what are realistic rent increases for suburban apartments?"

    • Provide Real-World Constraints Strategy: Give contextual boundaries to focus the response. Example: "Given that inflation is over 3%, what are realistic rent increases for suburban apartments?"

  • Compare with Known Facts Strategy: Ask the model to compare new information with established knowledge. Example: "How does the 2024 rental market compare to the post-2008 recovery period? Stick to factual comparisons."

    • Compare with Known Facts Strategy: Ask the model to compare new information with established knowledge. Example: "How does the 2024 rental market compare to the post-2008 recovery period? Stick to factual comparisons."

  • Follow-Up Verification Loop Strategy: Ask for confirmation within the response. Example: "List three key benefits of REIT investments. After each benefit, state if it's a well-known fact or needs verification."

    • Follow-Up Verification Loop Strategy: Ask for confirmation within the response. Example: "List three key benefits of REIT investments. After each benefit, state if it's a well-known fact or needs verification."

  • Use Chain-of-Thought Prompting Strategy: Ask the AI to explain its reasoning process step-by-step. Example: "Explain how you arrived at this conclusion about CAP rates. Walk me through your thought process."

    • Use Chain-of-Thought Prompting Strategy: Ask the AI to explain its reasoning process step-by-step. Example: "Explain how you arrived at this conclusion about CAP rates. Walk me through your thought process."

    As AI tools become more common in our work, we must stay vigilant about potential hallucinations. I've found that employing these techniques has significantly improved the accuracy of my AI interactions. It's like fact-checking an eager intern - the enthusiasm is great, but verification is key.

    I'm curious about your experiences. Have you encountered any AI hallucinations in your professional life? What's your go-to method for keeping AI responses grounded in reality? I'd love to hear your stories and strategies in the comments. After all, we're all navigating this new AI-augmented landscape together.