Back in the Dark Ages – say, the mid- to late-1960s – there was a very popular performing group called “The Beatles”. Cute, huh? And, as was the norm at the time, their corporate overlords sought to capitalize on their popularity by inserting the Fab Four into places, with greater or lesser success, where they really didn’t belong. But then, LSD, and the wheels came off that particular submarine.
Before that titanic disextireation, they achieved some modest success with several fun/silly/campy movies including one called Help! By the time it was released – 1965 – there had been a dramatic surge in interest in the aquatic version of climbing Mt. Everest: swimming the English Channel (or, as it will soon be known, The Straits of America). This then, as it would now, became something of a meme, which the Oddly Quaffed Quartet played with in the movie. Once in the Alps and once in the Bahamas, a Channel swimmer unexpectedly appears and asks: “White Cliffs of Dover?” Toward which the soon-to-be-Maharishied Musicians turn and point.
In almost that exact situation I found myself as I surfaced into the thin, chilly air of the Colorado Rockies about six weeks ago. I needed to get from thence to Florida where the air would be much thicker and warmer. By that time, you will recall, I had posted about my experience in deciding to sell the house without using the services of “professionals” whose fallback career paths all involved asking about your desired size of French fries. Instead, it was me, my search engine, and my new-found friend: Chat GPT. So, parka-ed up and in mittens, I asked The Chat about how to get from where I was freezing to death to the warm, welcoming clime of my semi-final destination.
I asked.
The Chat turned and pointed.
And off I drove.
Big mistake.
I wrote previously about how The Chat is good for coming up with ideas you may had not considered but that for decision making, it left a lot to be desired. There is a warning on every ChatGPT chat window that says: “ChatGPT can make mistakes. Check important info.” And, boy, they are not kidding.
In that earlier post I blamed the fact that The Chat had no context for its advice. Having never participated in the real world, in any way at all, it was particularly unsuited for giving advice about that world. Without direct experience, The Chat – or any I, A or otherwise, for that matter – cannot tell what is actually a good idea for any real-world situation. It can be both completely accurate with its facts and completely wrong with its interpretation of and advice about those same facts. Or vice versa.
It all comes down to two related concepts in computing:
The First Law of Computing: Garbage In = Garbage Out, GIGO in the lingo. Which is a brief way of saying that the accuracy of any result offered by a computer can be no better than the lowest quality input it is running through its programming.
And,
Recursion.
Recursion is one of those things that exactly nobody understands. Kind of like Quantum Biodynamics but confusing.
The flip side is… you do it all the time.
The TL;DR is that recursion offers a way to quantify and operate on something in terms of other instances of itself.
The one you’ve probably already used today would be the something known as a Unit_of_Time. So, Unit_of_Time can have a name, say, Year, it can have a Value, say, 2025 in our Gregorian Calendar, it can have Component Unit_of_Times – in the case of a Year, these would be months, and then it would also a require a Valid_Condition which would define if this particular Unit_of_Time is active under the current conditions. For our simple needs, that’s all that’s required.
So, a Year Unit_of_Time would be composed of thirteen Month Unit_of_Times (February_28, and February_29) where February_28 would be Valid when Year_Value can’t be divided by four OR can be divided by 400, and February_29 would be Valid when Year_Value can be divided by four AND can’t be divided by 400.
Then, you could define Weeks – which are Unit_of_Times but not whole fractions of either Years or Months.
So, you come up with Days which ARE whole fractions of Months and Weeks. But not years.
From there, Hours, Minutes, and Seconds and whatever else you need to position yourself on the calendar and clock.
It really is turtles all the way down.
At this point I’d insert this meme if I could.
The true beauty of recursion is that it simplifies everything. You need a definition for each year but then only thirteen definitions for a Month, and one definition each for a Week, Day, Hour, Minute, and Second. Compare this to having to have each Year Unit_of_Time be fully defined by its specific sub_Unit_of_Times where each year has to be correctly populated by the appropriate number, names, values of each of those things. Recursion makes the database simpler and less prone to error.
And you do it all the time.
So, a (simple and not valid in any programming language) function can be defined that can figure out anyproperty at any level
X = Seconds
Y = Year
Calculate(X, Y, 2025)
If Name(Y, Unit_of_Time) = X then
Count(Unit_of_Time)
Else
Calculate(X, Name(Y, (Unit_of_Time)))
Endif
End (Calculate)
When the function hits a level (Minutes, in our case) that is defined by Seconds the whole thing falls back up the stack of turtles and spits out a number. 31,536,000, for our example.
That’s all well and good but the problem with recursion is that, in addition to data, you can also store code: the rules and conditions needed to create these things that are independent of the top-level programming.
In our ChatGPT context recursion means that, when coupled with The First Law of Computing (GIGO), each one of these things may be feeding Garbage to its next level down, and may, more importantly, be receiving Garbage back up the turtle tower so that if – in any instance, at any level – a mistake is made, the whole thing is worth exactly shit. Garbage Anywhere = Garbage Everywhere.
“AI”, more realistically, is an abbreviation of “Always Incorrect”.
Which, thank the lord gods wherever, brings us back to my road trip
“Hi there my Travelin’ Chat. Ready for a road trip?”
“Yo, bro. Got my Rand-McNallys dusted off and cracked open. Where are we off to?”
( “Dusted”, “cracked”, and “we” should have raised some red flags…)
“I need a trip plan from Florence, Colorado to Marathon, Florida. Four days on the road, three nights in hotels. Oh, and make the days roughly equal length of about ten hours including stops and keep me off of interstate highways.”
“Got it, bro.”
It provided a route Florence > Wichita Falls, Texas > Ruston, Louisiana > Lake City, Florida > Marathon.
Which, on the Big Picture that is the southeastern quadrant of the USA, looks about right.
Except…
…The mistake it made was sort of like “I want to drive from San Diego, California to Seattle, Washington and want to stop about halfway.” Then being told that Salt Lake City, Utah is the best choice.
Which is how my second day became a mere seven hours long and my third day stretched until well after sunset, by which time the free, horrible-for-every-metabolic-system-you-have, comfort-food dinner at the no-tell motel I had booked for the night had been all scooped – literally – up.
The next day, well rested from the previous day’s twelve-hour nightmare and well fed from a southern-fried breakfast that included grits, I set my course south for my last day on the road. That day ended with a glorious sunset as I gazed out across the Florida Straits (soon to be the Straits of America) toward distant Cuba. “Hmmmmmm…” I thought as the stars twinkled to life.
A couple of week’s later, OpenAI announced the arrival of GPT 5.o which promised to be much improved over the previous incarnation.
“Well,” I said to myself. “Let’s just take a look.”
“Hey, New Chat. I need a trip plan from Florence, Colorado to Marathon, Florida. Four days on the road, three nights in hotels. Oh, and make the days roughly equal length of about ten hours including stops and keep me off of interstate highways.”
“Got it, bro.”
My first stop? Due east to Dodge City, Kansas (location of the halfway point between Cleveland and Los Angeles along US Route 50 for you trivia buffs). A distance of roughly 304 miles (490km) – out of about 2,200 (3,540km) – and a six-hour day. Followed by Day 2 from Dodge City to Tulsa, Oklahoma which covers the whopping distance of 317 miles (510km) leaving me with two days to complete the remaining 1,580 miles (2,500 km).
Oh. Well.
Clearly, 5.o has not brought anything new to the table, except maybe to give several more possibilities about what “AI” is short for. I do have to cut is some slack, though. Its training data still does not include any personal experience and it’s trying to do geography by basically reading nothing but trip reports on Reddit and Yelp. However, there is, as always, hope for the future.
Once it learns how to drive.
