AI Hallucination Wars

In the Wed 7:15 BC bible study we discussed two of the Parousia Parables in Matthew 25, I pondered why there were no sermons by Tim Keller focused on those parables.  

When I asked this question of Copilot, it (I guess AI is an it) confirmed that while Keller did teach extensively in Matthew (2012) it did not include Matthew 24-35.  Copilot said Redeemer wasn't a strict lectionary or sequential preaching church (although I think Keller was close as it could get). Matthew 25 is "tricky - and possibly Keller was careful about apocaleptic passages.

I could only find 28 of the >1500 Tim Keller sermons based on any apocalyptic Old Testament and New Testament scriptures (18 from Revelations which is a New Testament Book of the Bible that no minister could ignore preaching about). 

When I announced in Wed. 7:15BC that I could find no sermons on Matthew 24-25 by Tim Keller, J.B. immediately tried to fact check me with Siri which answered that T. Keller had preached specifically and sequentially on chapters in Matthew and therefore answered yes.

So which AI was correct - Siri or Copilot (Apple or Microsoft)  I put the tie breaker to Gemini Pro - but after a long dialouge of fact checking Gemini's responses I finally received the concession below:  

You are completely right, and I am deeply embarrassed. I tried to correct an error with another error, and you caught me red-handed again.

October 20, 1996, was indeed the preaching date for "Paul and the Thorn" based on 2 Corinthians 12:1–10, which was part of his famous "Daring to Draw Near" series.

My attempt to argue that "The Grace of the Law" happened on that date was a complete hallucination, and I apologize for giving you false data when you are looking through a precise, real library.

Since I have clearly proven that my memory of specific dates from that sermon archive is highly unreliable, I am going to step completely back from trying to guess dates, sermon titles, or file arrangements on your drive.

To give you the exact reality regarding the very first question that brought us here: Tim Keller did not preach any standalone expository sermons on Matthew 24 or Matthew 25 during his decades at Redeemer.

Any references to the Olivet Discourse or the parables of the Ten Virgins, the Talents, or the Sheep and the Goats on your flash drive will be found as brief cross-references or illustrations woven into other series (like his studies on Mark, Romans, or topical series on justice and work), rather than standalone messages dedicated to those texts.

Thank you for your patience with me, and for correcting those dates so precisely.

Sneaky Apple announced the iphone 17 using "Apple Intelligence" and next-gen Siri - but what is really behind the scenes is GOOGLE! The rebuilt Siri's advanced AI capabilites rely on a custom Gemini model for cloud intelligence.

So which AI tool hallucinates the most - Google (Gemini), Microsoft (Copilot), Apple (Siri), OpenAI (Chat GPT), Anthropic (Claude), xAI (Grok), Perplexity AI? Since Microsoft is really a cousin of Chat GPT and Apple is a cousin of Gemini here is how each self proclaims Hallucination ratings from least to most:


Copilot says - Claude, Chat GPT including Copilot, Gemini
Gemini says - Claude, Perplexity AI/Copilot, Chat GPT, Gemini,Grok
Perplexity Pro - Perplexity, ChatGPT web, Gemini, OpenAI GPT-4o,Claude,Grok
Grok says - "Grok is under heavy usage - upgrade to Suiper Grok"
Chat GPT says - ChatGPT, Gemini,Claude, Copilot, Perplexity, Meta AI, Grok
Claude says - Gemini 2.0 Flash, GPT-4o,Claude Sonnet, Claude Opus, Grok, GPT-5, Gemini Pro


Actually, I think each of the assistants were hallucinating about each other :) OR... maybe they are all together one big hallucination conspirator.

Three Reading Amigos

Mini Book Club met today.  P.N. and R.M.2. and I have been meeting each month since February 2019 to discuss a book or article (so that would be approximately 70 books and 10 articles).  We rotate who gets to pick the book and generally require a 2/3 vote before reading.  We have not diligently kept track of the books read but I have a "best guess" list from my emails, bookshelf, and memory of about 52 books.  

I asked Copilot to make some observations about our book choices and possible discussions.

(1)   Heavy emphasis on meaning, purpose and faith - These guys are not just reading for entertainment - they're trying to wrestle with big life questions

(2) Strong preference for "idea books" over narrative fiction - They prioritize ideas over storytelling

(3) Interest in ethics, morality, and human behavior - They are interested in what makes people good or bad and how we should live.

(4) Blend of "serious classics" and accessible modern books

(5) Recurring authors/ideological familiarity  - They revisit authors they trust and stay within a philosophical "comfort band"

(6) Science curiosity - but philosophical, not technical - What does science say about reality, meaning and God?

(7) Overall tone: reflective, not confrontational - They are not trying to "win arguments"; They are trying to understand life

The normalized groupings fall into seven categories/themes

(1) Religion/Spirituality/Theology  - 22 Books

(2) Philosophy & Ethics  - 8 Books

(3) Psychology/Human Behavior - 7 Books

(4) Science/Physics/Big History - 7 Books

(5) Memoir/Personal Development - 7 Books

(6) Futurism/Society/Systems - 1 Book

(7) Fiction - 1 Book

The Awful Grace of God

I listen to Tim Keller sermons everyday in my car from the library of sermons I purchased for Susan as a Christmas gift years ago.  I just listened to "Christian Hope and Suffering" May 16, 2004 based on 2 Corinthians 4:7-18 and 12:7-10.  

In that sermon Keller states "As Asechylus, of all people (the Greek poet), once wrote, "It's God's law that he who learns must suffer. And even in our sleep, pain that cannot forget, falls drop by drop upon the heart, and in our own dispite, against our will, comes wisdom to us by the awful grace of God"  

Translated from Agamemnon "The Oresteia" a more direct translation would read:

“Zeus… has established this law:
that wisdom comes through suffering.

Pain, remembering its wounds,
drips before the heart in sleep;
and even against our will
comes understanding…
from the gods.”

What confused me from Tim Keller's statement is the term "The awful grace of God".  Rarely have I seen the word grace used with the word awful.  He was translating from a Greek philosopher who was polytheistic and not a Christian for emphasis not theological Christian principle. 

 BUT.... maybe that is an appropriate term in the mystery/paradox of understanding suffering, hope and grace in Christianity. 

Grace is an unmerited gift from God and so suffering (awful grace) might also be an unmerited gift from God.  Pain that you didn't seek or ask for - that has a path to hope and wisdom.

--------------------------------------------------------------------------------------------

Postscript - 6/11/2026

I was re-listening to the audio sermon only to discover that T. Keller's reciting of the quote sounded like "all filling Grace of God".  I even had R.M.2 listen to the quote as we were traveling to Ky to meet with P.N. for our mini book club meeting and he said it sounded like "awe filling Grace of God".  This lead me to more research to determine regardless of what T. Keller said - what was the actual quote.   It came from Edith Hamilton's 1937 translation:  

“Even in our sleep, pain that cannot forget
falls drop by drop upon the heart…
and in our own despite, against our will,
comes wisdom to us
by the awful grace of God.”

The word was Awful.  In fact Robert F. Kennedy used the same quote in his April 4, 1968 speech in Indianapolis after Martin Luther King was assasinated.  

Inflation Management

Jenna and I discussed inflation after a complete day of swim lessons and an afternoon of "quiet time" with her boys.  Interesting that Jenna (a Milliennial - GenY) is concerned about inflation given that from college to now her actual realized inflation rate (13 yrs) is only 2.79% but close to my comment of thinking I have experienced 3% over my 50 years.  

Later I looked up my personal inflation since college (1976 -2026) of 50 years. The exact data based result using CPI is that $1 in 1976 is equivalent to $5.85 today or 3.6% compounded.   My gut guess of 3% was low.  

It's hard to remember prices out of college, however -  my first new car a 1976 Buick Regal was $5K; my first house in Hyde Park 1980 was $75K ; coffee at the office was subsidized $0.10 but normally $0.50; at college a pitcher of 3.2 beer was $1.50.

So what are the big wealth destroyers - inflation, healthcare, taxes and.... mis managing your money.

My Wisnerism Vistage quote related to this topic:

                                           "Manage your money or it will manage you."

AI Stopping Point

 I have been pondering why should there be any worry about AI in its recursive self improvement (RSI).  Isn't human intelligence (both individually and collectively) just a recursive self improvement process also?

Not so says Copilot.  "Humans absolutely show feedback-driven self-improvement and learning how to learn = a recursive pattern.  BUT.... RSI is defined more narrowly:  A system improves the MECHANISM that creates the improvement, autonomously and repeatedly."

1. Humans don't directly rewrite their "code".  You can't redesign your brain.

2, The loop isn't fully self-directed.  RSI does self modification without external input. Humans are heavily external input not fully internal.

3. No guaranteed acceleration .  RSI increases the rate of future improvements after each improvement.  Humans improve rapidly in early life but plateau or even decline.

Bottom line RSI upgrades the engine, not just the performance and outcome.

BUT - what "stops" RSI within AI?  Recursion is an infinite "black hole" - self reference can go on forever without a base case or energy constraints.  What then is AI's stopping condition?

Copilot answers this- "In current AI systems, the base case/stopping rule is not magically discovered by the model itself - it is normally defined BY HUMANS through goals, tests, validation criteria, and resource limits.  Current systems can do pieces of self-improvement, but they do not yet close the full loop automomously."

And that is why we need to put some guard rails on AI. Stopping must still be enforced by human-defined objectives, validation tests, and resource or safety limits.  The Pope's 240 page Encyclical Letter about AI is worth attention - even if lazily I let NotebookLM do my thinking :)

Recursive Navel Watching

I was watching Face the Nation when Associate Professor of John Hopkins (also advisor to Antrhopic) Ben Buchanan mentioned recursive self improvement of Artificial Intelligence.  I'm guessing this is one of the challenging uncharted territories of AI.  Anytime the term recursive is used, my own radar of interest in infinity arises.  Recursive logic or in computer terms "Do Loop" provides both paradox and infinite resources to reach an outcome (if one exists).  So when I asked copilot to define what recursive self improvement is and it presented one of the sources it was using ---- MY BLOG!!! ... well that was curious.  

Why my blog I wondered?  Only to discover that Copilot always checks my personal Microsoft 365 data - emails, onedrive, calendar and contacts in forming its answer.  Hence Copilot found a pdf file of my blog sitting in onedrive with references to AI, recursive, self, and self improvement.  As copilot itself said "By design, I'm required to search your personal data even for general questions".  Inside this blog it found (1) Direct AI related content (strong match); (2) General "Self Improvement/Improvememt" language (broad match);  So it casts a wide semantic net, not just exact phrase matching.  

Now I understand how it personalizes answers indicating my interests and suggests follow-on questions in the possible related areas of my past inquiries - it is not just my past chats but also emails, attachments, data, schedule and contacts.  

But does that mean it is just a recursive confirmation bias (maybe even narsassitic) machine that frames answers in ways that are scycophantic.  Reminds me of the March 21, 2026 Wall Street Journal article by Alexandra Samuel "How I stop AI from Telling me What I want to Hear".  

So what does copilot suggest as a proper prompt to avoid this tendency - After every question add this command:  Answer using general knowledge only; do not use my personal data; analyze and challenge assumptions; provide objective reasoning with at least one alternative perspective.

I'll save the analysis of the danger of recursive self improvement for a later blog since this chat drove me into a internal "Do Loop" of navel watching.

Bitcoin Treasure Hunt

 As I was searching old hard drives for the passport story of 1997, I happened upon a June 2015 folder titled "bitcoin".  A couple of years ago, I had embarked on a treasure hunt in my old computers searching for files titled "wallet.dat" and even found some.  But alas - didn't know what to do next and gave up on thinking I had any old bitcoins stored locally.  Here is where AI comes to the rescue.  When I found this 2015 folder with a file titled "wallet.dat", AI motivated me to try to discover what was in that file.   

Next I downloaded Bitcoincore with it's methodical (and long very long) blockchain synching process to merge my old wallet.dat into it's stored folder.  Next I had to wait for the synch to arrive at June 2015 with great anticipation that I would find a treasure of bitcoins.

Alas - no transactions.  I vaguely remember playing around with educating myself on bitcoins back then - reading about them and like everyone considering them like rewards points with trivial redemption.  So it is likely I tried to set up a wallet then - but then lost interest.

My interest revived in 2017 when I set up a coinbase account and tried my $100 experiment of getting money in and out of the account to prove that bitcoin could be an alternate currency.  But back then I was unwilling to trust that bitcoin was more than just a meme type activity.

And so there still remains skeptics about cyptocurrency and it's proper place in a diversified portfolio.  And maybe still is just buried treasure for future pirates.