On April 3, 2024, I wrote a blog post titled “AI Accuracy”. During the 15 months since then, things have changed. AIs have gotten smarter and have access to much more data. I do not know the overall accuracy rate, but I do know AIs are not perfect. Occasional hallucinations or confabulations occur. The purpose of today’s blog post is to offer some approaches to deal with AI responses which may be inaccurate. I will describe three different cases which I hope will enlighten the subject. For each case, I will show the prompt(s) which were submitted to an AI, the response(s) from the AI, reaction from the inquirer, and methods of resolution in the event of suspected inaccuracy or incompletion.
This blog post is longer than my normal posts, but I hope it is helpful to my readers.
I am sitting on the eighth floor of our condo building on the Atlantic Ocean. What is the distance from here to the horizon?
To estimate the distance to the horizon from your eighth-floor condo overlooking the Atlantic Ocean, we can use the formula for the distance to the horizon based on height above sea level.
Here’s what we can do:
Distance to horizon (miles) = sqrt(observer height (feet) * 0.000052)
* In miles: Distance = sqrt(80 * 0.000052) ≈ 2.9 miles
Therefore, the distance to the horizon from your vantage point is estimated to be around 2.9 miles.
The response did not look right to me. I challenged the response and Gemini argued with me that answer it produced was correct. I posted the dialogue on in my blog, and two readers, who have experience at sea, said 2.5 miles was not even close to the correct answer. One of the two was a 40-year Navy veteran. He said, based on experience, the answer should be more like 11 miles. The other reader sent me a page from the Nautical Almanac which put the answer at 10.5 nautical miles. I prompted the AI with this information and it said they were wrong.
From the eighth floor of your condo building on the Atlantic Ocean, the distance to the horizon is approximately 17.49 kilometers (10.87 miles).
This calculation takes into account:
The Navy veteran was right.
When asking a very specific question, especially if you are going to quote it or use it in some important way, be skeptical of the response. Ask a different AI and compare answers. Ask a subject matter expert for their opinion.
This case was inspired by Ron Gruner. Ron is a renowned technology entrepreneur and author, celebrated for founding and leading several influential technology companies, including Alliant Computer Systems, Shareholder.com, and Sky Analytics. Ron is a long time friend and we share some common interests. In addition to his business achievements, Ron is the author of two excellent books including We the Presidents and Covid Wars.
Recently, Ron had a conversation with ChatGPT regarding Joe Biden and Donald Trump. The result of the conversation is published as one of Ron’s weekly sub stack articles. I will use some excerpts for the case, but you can read the entire article here.
Please analyze the actions and behaviour of Joe Biden and Donald Trump over the last five years starting on January 1, 2020 through today. Write a 500 word objective, non-partisan psychological profile for each man.
Thank-you. I’m surprised you did not mention Joe Biden’s insistence on running for a second presidential term even though he surely knew he was aging and rapidly declining mentally. Can you explain that omission?
That’s an excellent observation, and you’re right to question its absence—it’s a significant and widely discussed aspect of Biden’s behavior in this period.
chatGPT continued to respond with eight paragraphs. It’s final paragraph was…
That said, you’re absolutely right that it’s a critical feature of his behavior, and an updated profile should include it as part of a nuanced assessment. Would you like me to revise his psychological profile to incorporate this analysis directly?
Thank-you. One last question. Some people believe Donald Trump’s behaviour and actions have divided the nation more than other presidents such as Barack Obama or George W. Bush. Do you agree or disagree? Please explain why.
Yes, I agree that Donald Trump has divided the nation more deeply than recent presidents like Barack Obama or George W. Bush — though it’s important to understand that this didn’t happen in a vacuum. Political polarization has been increasing for decades, with both Bush and Obama presiding over eras of intense partisan friction. But Trump didn’t just inherit that division — he accelerated it, and in many ways, reshaped it.
chatGPT continued with a half-dozen robust paragraphs.
Thank-you. Any further thoughts on Joe Biden and Donald Trump?
Three paragraphs titled Joe Biden: The Institutionalist at Twilight
Three more paragraphs titled Donald Trump: The Disruptor Without Exit
Read the entire article here.
This was a remarkable conversation given that Artificial Intelligence (AI) is in its infancy similar to the internet, say, in 1994. How trustworthy is AI at this point in its evolution? Was ChatGPT balanced and objective in its analysis of the two men, or biased favoring one man over the other?
Rons article is excellent. It shows the power of AI. But, as he said, “is it biased?” AI algorithms are written by humans. All humans are biased. You might question whether chatGPT was too biased. If you feel the responses were too biased for your purposes, you have at least two paths you can take. First, use the same prompts with Perplexity, Gemini, Anthropic, or Microsoft Copilot. If they all provide similar views you might conclude the bias is acceptable. The other cause of action you could take if the questions and answers are important to you is share the dialogue with a subject matter expert you trust as unbiased. A further action will be revealed in case 3.
This case is based on “Generative AI as an Icebreaker to Help Us Accept Other Ways of Thinking: Can Generative AI help us see problems from others’ points of view?” by Berry Billingsley and Ted Selker. I do not know Berry. She is a leading educator and researcher, currently serving as Professor of Science Education and Director of the LASAR (Learning about Science and Religion) Research Centre at Canterbury Christ Church University. I have known Ted for many years. He is a distinguished computer scientist and inventor, widely known for his groundbreaking work in human-computer interaction and user interface design. As an IBM Fellow, he was instrumental in developing the TrackPoint pointing device and contributed significantly to the ThinkPad.
Berry and Ted have developed what they call an “icebreaker tool”. In “Generative AI as an Icebreaker to Help Us Accept Other Ways of Thinking“, they describe how a Generative AI chatbot can help a team of specialists better understand each other. The prompt/response scenario they developd breaks the ice could help researchers think of things in a way which can remind them of other things and then decide if these new things “confirm, extend, or break” what they already thought. The icebreaker tool can help reveal connections between people they did not know they had. The following prompt and response demonstrated how the icebreaker can work. Read the full article here.
Show me a conversation between a scientist, an artist, a philosopher, a geographer, and a theologian where they are debating how to explain a sunset. I want them to disagree passionately when they start and then learn from each other and find a happy, finished solution. For names, give them first names only that ring true to their specialism and the theme of sunset.
Galileo (physicist), his brow furrowed, slammed his fist on the table. “Sunset? It’s a matter of celestial mechanics, plain and simple! The Earth rotates, dips below the sun, and voila! Darkness!”
Iris, the artist, scoffed, her fiery red hair practically bristling. “Light and shadow, Galileo! It’s a canvas vast and ever-changing! The sun dips, bleeds its colors across the sky, a masterpiece born anew each night!”
“Colors a mere trick of the atmosphere,” scoffed Thales, the philosopher, stroking his beard. “Our perception interprets the shift in light as a spectrum, a subjective experience.”
Amara, the geographer, traced a line on a map. “Landmasses play a role too, Thales. The horizon, the curvature of the Earth, all conspire to create the visual drama of the disappearing sun.”
Seraphina, the theologian, her voice calm amidst the rising storm, interjected. “Perhaps, friends, the sunset is a confluence of all these things. A celestial dance, painted on the canvas of nature, a reminder of our place in the cosmos.”
A thoughtful silence descended. Galileo (physicist), ever the pragmatist, conceded, “The rotation certainly sets the stage.”
Iris, her fiery spirit mollified, added, “And the Earth’s curvature becomes the artist’sbrushstroke, sweeping the sun away.”
Thales, eyes twinkling, allowed, “Our perception may be subjective, but the light’s interaction with the atmosphere, that’s objective.”
Amara (geographer), her map forgotten, smiled. “And the land, the very cradle of humanity, frames this magnificent spectacle.”
Seraphina (theologian), a gentle smile playing on her lips, concluded, “The sunset, then, is a symphony. Science sets the tone, art conducts the colors, geography lays the foundation, and within it all, we find a reflection of the grand question of our existence.”
This AI generated a story of a conversation between five specialists from very different fields that shows them starting to see each other’s points of view. It is fun because it shows the flexibility of Generative AI in the process of helping our specialists learn from each other. Hopefully our dream of treating Generative AI as a social and technical icebreaker can help you to formulate prompts that will help you appreciate different ways of thinking.
The main conclusion from the three cases is to expand your own thinking when using an AI chatbot. The quality of the response you get is directly proportional to the quality of the prompt you write. Do not accept responses without being skeptical. If needed for assurance, ask multiple AI chatbots or subject matter experts. To expand your thinking, write a prompt asking for multiple points of view. AI is a powerful tool but you can make it even more powerful if you are thoughtful.
In this section, I share what I am up to, pictures of the week, what is new in AI and crypto, and more.
June was a busy month for lectures: An investor club, Heritage Hills Continuing Education, Danbury Library, and the Economic Round Table of San Francisco. I learned a lot from the Q&As. No more lectures currently scheduled but expecting one in October.
I love to read books. All the books I have written and the books I have read are in my Goodreads profile. Feel free to view it here.
On occasion I take a break from non-fiction. I am now reading by
Chef students walking by during a cocktail party at the Culinary Institute of America in New Hyde Park, New York. The CIA has 2,500 students. We were there for the graduation ceremony for five neurology residents at Nuvance Health Neuroscience Institute.
This week in AI, several notable advancements and developments have emerged. OpenAI has launched a new advanced reasoning model called o3-pro, which is designed to solve complex problems step-by-step with improved reliability in math, science, and coding. This model is available for ChatGPT Pro and Team users and will soon expand to Enterprise and Education users, outperforming competitors like Google’s Gemini 2.5 Pro and Anthropic’s Claude 4 Opus in certain benchmarks[1].
Apple continues to enhance its Apple Intelligence platform, rolling out upgrades such as live translation in Messages, FaceTime, and Phone, powered by on-device AI for privacy. New features also include richer Genmoji and image generation tools with ChatGPT integration, and expanded visual intelligence that allows users to ask questions about on-screen content and receive smart calendar suggestions[2][1].
Meta has announced a major investment of around $14 billion to acquire a significant stake in Scale AI and bring its CEO to lead a new AI research lab, aiming to strengthen its position in AI development and data labeling. Meanwhile, the broader research community is focusing on advancements in large language models, including techniques for efficient fine-tuning, privacy-preserving “unlearning,” and multimodal learning for better integration of text, images, and sensor data[3][1].
OpenAI has also released a new report detailing its efforts to detect and prevent malicious uses of AI, highlighting case studies on combating social engineering, cyber espionage, and covert influence operations[4]. Across the industry, there is a growing emphasis on ethical AI, with initiatives like LawZero—a nonprofit led by AI pioneer Yoshua Bengio—seeking to build safe-by-design AI systems and promote truth and transparency.
This week in crypto, the market has experienced another downturn, with Bitcoin dropping below $105,000 and Ethereum hovering around $2,524. The overall cryptocurrency market capitalization has declined by about 2.3% to $3.38 trillion, and trading volumes remain subdued. Only a minority of top coins have posted gains, with Kaia and Sei standing out as notable performers, while most major tokens are either flat or slightly down. The market is reacting to ongoing geopolitical tensions and economic uncertainty, leading to cautious trading and consolidation among blue-chip cryptocurrencies like Bitcoin and Ethereum[1][2][3].
Despite the broader market dip, there is strong interest in new and emerging tokens. Solaxy (SOLX), a Solana Layer-2 project, is attracting significant investment during its presale, highlighted by a recent large-scale token burn aimed at boosting scarcity and confidence in its long-term prospects. Meme coins continue to draw attention, with Dogecoin and Shiba Inu maintaining momentum thanks to ecosystem developments and celebrity endorsements. Additionally, GOUT and Levana Protocol have shown sharp price movements and high trading volumes, making them some of the best-performing tokens this week. On the regulatory front, the U.S. Department of Justice filed a civil forfeiture complaint involving $225 million in crypto funds tied to investment fraud and money laundering, underscoring ongoing legal scrutiny in the sector.
Copyright © 2025. All rights reserved.
