top of page

Is it accurate? is a powerful use case. Let's talk.

“Is it accurate?” is a powerful use case for an artificially intelligent system. Read to the end. If you are in a business where you must comply with standards, from tax law to medical device regulations or building codes, let’s talk. I have thoughts about helping you develop an intelligent agent that enhances your productivity and makes you more competitive.


In The Intelligence Tsunami (, I tell this story of breakfast with two friends. 

"John drove his car to his local airport to fly to a meeting in Albuquerque, NM. At that airport, he rented a pickup truck, which was unusual for him. Later, when loading an item into the truck’s bed, he noticed that the open tailgate blocked the truck’s backup camera. A week later, he had breakfast with me and another friend, Walter, who had recently crashed his car in an accident. By chance, Walter also rented a pickup truck while his car was being repaired. With no knowledge that John had been in Albuquerque, Walter said his truck’s backup camera didn’t work properly. John asked the question that immediately popped into his mind, “Was the truck’s tailgate down?” Walter perked up, “Yes, it was.” “That’s what blocked the camera,” John said."

As that happened at breakfast, I was amazed that John’s brain could pull that esoteric piece of information from his memory and apply it to Walter’s situation to diagnose the problem with Walter’s truck’s backup camera. When John noticed his backup camera was blocked, he had no reason to believe that nugget of information would ever be useful again, yet he could retrieve it when it was. 

Something similar happened while writing the book. I would open ChatGPT and Google Gemini, write the word “Accurate?” in the prompt box, and then cut and paste a section of text from the book. Both large language models (LLMs) would analyze the text, tell me whether they thought it was accurate, provide bullet points describing how they came to that conclusion, and provide suggestions if they felt something wasn’t as accurate as it could be. Try that yourself with a piece of text you wrote. You’ll be impressed. You’ll also find that ChatGPT and Google Gemini have different personalities in how they analyze text and provide feedback. 

From 2000 to 2005, I was the point person on Wall Street as an executive at the public company KEMET, which had been a subsidiary of Union Carbide. In 1984, a massive gas leak occurred at a Union Carbide pesticide plant in Bhopal, India, exposing over 500,000 people to highly toxic gas and killing over 2,000. The story I heard from the KEMET CEO and repeated to Wall Street analysts many times was that Union Carbide went bankrupt and sold off its non-chemical-related businesses. I wrote a story for the book about how KEMET's executives all made small fortunes leading the company’s buyout from Union Carbide.  

I wrote the word “Accurate?” in the ChatGPT and Google Gemini prompt boxes and cut and pasted the text of that story into the LLMs. Amazingly, ChatGPT answered that the story was not quite accurate. Union Carbide had been restructured, but it had not gone bankrupt. I was stunned. That’s not the story I heard from an authority, KEMET’s CEO. That led me to research the issue more. ChatGPT was right! Think about that amazing ability of ChatGPT to pull that esoteric piece of information from its memory and apply it to my situation. ChatGPT had been trained on public Internet data and had stored that nugget of information to retrieve at precisely the right time. This is similar to how John’s brain assessed Walter’s situation. The backup camera wasn’t broken; the tailgate was down.


In a former life many moons ago, I was a tax accountant with the international accounting firm KPMG. I’d write memos for clients analyzing the tax consequences of situations they found themselves in. For example, I analyzed the tax treatment of Fripp Island’s bankruptcy on my savings and loan clients.


Imagine if I could cut and paste the text into an intelligent agent, which analyzed my work using the Internal Revenue Code, its regulations, and court cases. The agent would provide a report documenting how it concluded I was accurate and, even more powerful, provide suggestions for improving my conclusion. KPMG was by far the leader in our market, serving about two-thirds of the savings and loans in the country. Rather than the intelligent agent being trained only on public internet data, imagine the agent trained with proprietary data from the thousands of memoranda Peat Marwick professionals wrote about various tax consequences of savings and loans. This could significantly increase the productivity of KPMG professionals and enhance the firm’s distinctive competitive advantage in this market. 

If you must comply with government or industry standards, let's talk.

John Warner


31 views0 comments


Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page