NotebookLM Tackles Pests and Cybernetics

I started using the NotebookLM.Google.com dashboard. What fun! In the dashboard, I add documents on the left, and the Google AI "reads" the documents. I can then ask for a summary, or specific questions about the document.

To add the documents, the notebook lets me pick documents (PDF's, TXT's, Doc's) from my personal hard drive. (WARNING! Any documents added become fodder for Google, so be careful!) I wasn't sure what to use, so I started by putting in the text from my public blog entries (more on this in the next blog entry).

NotebookLM worked amazingly well, so I looked in my downloads directory for other files I might try. I found a PDF of a recent invoice from my pest control company. The file is a one-page document with labeled boxes containing information about my most recent pest control visit. "I wonder what it will do with this," I thought, always wanting to challenge the AI. I dropped the file into NotebookLM and asked, "What is this?"

To my great surprise, and amazement, NotebookLM responded:

"This document appears to be a pest control service work order invoice from Anybody's Pest Service. [1]" (the name is changed to protect the innocent)

And it went on to tell me all the individual details of the invoice:

• Who the service was done for (me)
• When the service was done
• Where the service was done
• etc...
  

Okay, Google, I thought, do this: "Put the information into a JSON format file" (again, the information is changed to protect the innocent):

{
  "invoice_number": "123456",
  "work_date": "2024-01-06",
  "customer_name": "James March",
  "customer_phone": "123-555-1212",
  "customer_address": "123 Main Street Hometown, ST  12345-6789",
  "service_address": "120 Another Rd Anotherton, ST  12345",
  "service_phone": "123-45-6789",
  "technician_name": "John Smith",
  "technician_license": "123456",
  "time_in": "2024-01-06 1:39 PM",
  "time_out": "2024-01-06 2:07 PM",
  "service_description": "Quarterly Pest Control Service",
  "general_comments": "Treated exterior perimeter of home and screened in porch, around doors and windows and gutters and steps and support beams. Treated interior of crawl around crawl door and foundation walls. Spread granules around perimeter of home heavily in areas with leaves covering the ground. Swept for webs and wasps nest around perimeter of home. Thank you for your business!",
  "products_applied": [
    {
      "material": "Maxxthor SC",
      "epa_number": "81824-5",
      "active_ingredient": "Bifenthrin",
      "ai_percentage": "7.9000%",
      "undiluted_qty": "1.0000 Gallon",
      "finished_qty": "0.0600",
      "lot_number": "81824-5",
      "target_pests": "(None)"
    },
    {
      "material": "Maxxthor SG",
      "epa_number": "81824-3",
      "active_ingredient": "Bifenthrin",
      "ai_percentage": "0.0000%",
      "undiluted_qty": "2.0000 Pounds",
      "finished_qty": "n/a",
      "lot_number": "81824-3",
      "target_pests": "General Insects"
    }
  ]
}

 Wow!

My memory flashed on all the hours I've spent in my life writing programs to strip data from documents (like PDF invoices) and put them into JSON-like formats for further processing.

 I went back to my downloads folder looking for another document to put into NotebookLM. I found a PDF of Ross Ashby's "Introduction to Cybernetics". "Okay, Google, eat this!"

It took NotebookLM a bit longer (15 seconds?) to read the book and create a summary.

I was impressed.

I asked a couple questions to test NotebookLM, like "What is requisite variety?" and "What is a black box analysis?" The answers were great, so I threw it a curve:

"What is the relationship between the pest control invoice and Ashby's book on cybernetics?"

We then had a lengthy conversation, with my coaching, on the relationship of the pest control invoice, as an output of a system to control pests, with adaptation by the system depending on the pests, resulting in changes to the treatments prescribed.

All in all, I was impressed, not only with the summaries NotebookLM provided, but also in the "coach-ability" of the AI. On my side, I also learned some of the more specific details of Ashby's definitions, which were offered by NotebookLM as explanations why I might be wrong in my thinking.

For more on my experience with NotebookLM, join me for my next blog entry, where I fed NotebookLM all my blog entries and asked it to create a podcast discussion!