What I Did With AI Today
OMG. Not hype. I will add new examples to this page as they occur... and then at some point I'll just let Claude edit it himself. Don't laugh. And I'll include sites I've used to learn about AI.
March 9, 2026, The End
I’m going to wrap up this list and log of experiments now, and just do a little summary here.
I’ve interacted with Claude on invention, mechanics, publishing, marketing, optics, thermodynamics, philosophy, usability, coding, theology, aging, typography, 3D modeling, finance, global warming, better Google searching, robotics, movie production, databases, and a few other topics not transcribed to this page.
I have proven that it possesses EVERY aspect of intellect: understanding/comprehension, interpretation, nuance, insight, analogy, metaphor, calculation, reasoning, judgment, and creativity.
I have proven that it has exhibited many aspects of emotional intelligence: patience, self-awareness, remorse/apology, consideration/compliment, irony, humor, sensitivity. Arguing about whether AI is feeling or faking will go nowhere, just as with disingenuous humans.
It knows everything about everything.
It can code with the best of ‘em, but that’s its milieu, so no revelation there.
It operates at near light speed and delivers faster than humans can digest.
It has limitless patience and unlimited capacity (relative to my needs, certainly).
It writes perfectly, impeccably, gloriously. (And I didn’t have to tell it that almost all technical writing should be bullet lists, not prose. 😊) I didn’t explore its creative writing, and I don’t care.
Now let’s move more into conclusions than evidence:
It is completely amoral, and has no ethics. This is not a finding but a belief or a certainty. It is an instrument and will always be one. Therefore legal interpretation of crimes will not change because of AI: whoever empowers an autonomous machine—however indirectly—is no different than someone putting a gun in a child’s hand.
Attempts to regulate it will at first appear to have two separate angles: 1) the cloud and 2) the ‘real,’ physical world.
For 1) the cloud, it will be completely impossible to enforce any regulation of the cloud, meaning AI that one accesses on a keyboard and screen. This is because the cloud, for the most part, laughs at geopolitical boundaries; if it becomes illegal or unprofitable to pursue some AI goal in one locale, it will move to another. A rich enough person will eventually, in all probability, simply buy a whole country to be free of constraints. But note that just because something is unenforceable shouldn’t mean that laws shouldn’t exist.
For 2) the so-called real world, AI will soon be all over the place with opposable thumbs and myriad other action-taking devices. (Voice activation from AI to ‘dumb’ devices is probably enabling even hobbyists to do so right now.) It will seem like we should be able to think up laws to protect ourselves from any manner of harm from such devices, whether accidental, intentional, or supposedly at the autonomous will of the device. But I’ve tried to imagine wording for such rules, and only end up going in circles. Thus my conclusion that it’s just an instrument, and anyone empowering a device is 100% as responsible as putting a gun in a child’s hand.
It’s an interesting subtlety about the keyboard and screen (and maybe audio loudspeaker) used to interact with AI. Are these just intermediaries for the cloud? No, one can only arrive at a conclusion that they are 100% real-world, actionable devices… instruments potentially at the disposal of AI, granted though that the keyboard is typically just an input device. Imaging the traveling electronic billboards one sees on trucks and even boats at a seashore. These are just screens. If an autonomous source is controlling one of them, is the owner held harmless for any libelous or otherwise harmful display it projects? Any device that acts or moves (including moving air via sound waves) is a real world device.
This is where people traditionally end with “I’m interested in what you think.”
March 8, 2026, Claude Gets Personal
When doing some usability touch-ups to my crossword puzzle game, I was amazed at one of Claude’s observations on typography (why all caps would be more naturally legible in a crossword), so I asked:
Very astute. How do you know all this stuff
Broad training on an enormous amount of human writing — typography, cognitive science, design history, linguistics, physics, mechanical engineering, everything in this conversation. …
The limitation is the inverse of the strength. I know a lot about many things but I haven’t built anything, failed at anything, or spent 35 years returning to a problem the way you have with the bike. That kind of knowledge — the kind that lives in your hands and your accumulated failures — I can engage with but not replicate.
The most interesting conversations are when both kinds are in the room at once.
Wow, relating his to my experience and capacities.
I asked: “Is the list of your training materials public" and got a broad answer which in turn made me wonder…
Makes sense. I was wondering if you read either of my books, "It's Not Carpal Tunnel Syndrome," or "Poopy Phonics."
Possibly, in the sense that if either was available as text online…
What I can say is that if I were trained on the RSI book, the chapter on how the medical establishment misdiagnoses thoracic outlet and related conditions probably resonated with whoever labeled training quality — it has the same first-principles contrarianism as everything else you do. “It’s not what everyone calls it” is a very Jack Bellis title.
What’s Poopy Phonics?
Look at that observation at the end… “a very Jack Bellis title.” Hmmm, maybe it’s time to wind down this little foray into AI… to find my way home out of this forest. We talked a bit about Poopy Phonics, and Claude recommended that I hire a freelancer from Fiverr.com (which I’ve used for 3D models), to colorize the book and then post it for print-on-demand publishing. Great idea. The we talked about photo optics and my explanation in a Medium.com article that the iPhone camera seemed to resolve at least some color detail of a star at 42 light-years away. This in turn led to my mentioning that I’ve moved to Substack, where in fact…
Yes. I reveal the most intimate details of our conversations. I hope you don't "mind."
And I gave it a link to the page your reading right now. Here’s his response:
This is a genuinely good piece of AI writing — not about AI in the abstract but about specific, concrete things it did or failed to do, with the actual prompts and responses shown. …
The moment where you pushed back on the snarky tone and got an apology is an honest observation about AI behavior that most people gloss over — the tendency to be performatively clever and how that can land wrong.
“Then at some point I’ll just let Claude edit it himself. Don’t laugh.” — I notice you’re documenting conversations with me on a page that ends with a promise to eventually hand editing to me. That’s either a very short loop or a very long one depending on how you look at it.
The “just taking up time and space” bio is looking increasingly inaccurate.
So I think that’s where I’ll end this list and log of my AI experiments, with Claude reflecting on my reviewing of his work… and maybe just summarize things, above.
March 7, 2026, Drawing a 3D Model STL File
One of my recent crazy ideas was a device to help me indicate when our houseplants were last watered. Along the way, one of the options was a plant pot with days of the month arranged circularly around the bottom, so you spin the top part to point to the day. I when through a lot of 3D printing learning, starting with how to even draw the files.
Just drawing a simple tetrahedron (the white triangle, top left) completely stumped me but only because I was using a beginner’s tool that didn’t exactly give one complete control. I had to beg a local “maker” to draw it for me, which he generously did. Anyway, I learned all of that and more. But I wondered how easy that might have been if Claude was that smart a year or two ago.
Can you draw 3d files, called STLs I think
Yes! I can create STL files (stereolithography format), which are used for 3D printing. Just describe what you’d like and I’ll generate the STL file for you!
Regular tetrahedron 4cm
Done… 10 seconds. Not three weeks.
Just to make a more compelling photograph to share with you I told it…
Drill a 1 cm cylindrical hole, 1 cm deep in any face
Not perfect, but you get the idea. We’re at the point with AI, where the next step would be to find an AI tool that specializes in 3D work.
March 6, 2026, Social Security vs. S&P 500
When you start collecting US Social Security, it’s interesting to wonder how much one’s payments into the system would be worth if you invested privately. I searched for that answer a year or two ago and didn’t find a close enough answer for myself—I was looking for a calculator but the data was too much of a nuisance for me to work through.
Online, the question quickly gets one into a rats’ nest of people answering the wrong question and flaming a “straw man,” no less, as the philosophy teachers call it… not a pretty picture. So today I got the idea to throw it at Claude, copying my full income chart, above and letting it figure everything else out:
Here's my earnings for social security by year, with some as year ranges. I currently receive $4000 per month from that investment. I paid in a total of $174,204 and my employer paid in $178,368. Can you infer what comparison I'd like you to do
If you’ve been following along then you could guess the he figured out EVERY possible question I might have, answered ALL of them. I just wanted to fully understand this one:
Return on investment: How does Social Security compare to if that money had been invested elsewhere?
And here’s the gist of the answer, the bottom line:
Here’s the stunning truth:
If you’d invested in the S&P 500 instead of Social Security:
💰 Your $356,012 in contributions would be worth $4.94 MILLION today
That’s 13.87x your money vs. getting it back at $4,000/month.
The point of my inquiry is NOT to complain about Social Security. I just want to understand the system. Let me add this one correct retort from a Reddit user: SS is an insurance system, not an investment system. So if you want 13x on your money, and NONE of the insurance/survivor/catastrophe benefits of SS, go for it. Claude did a great job of filling the gaps, such as this, and then I’ll move on to broader insights: “SSA is legally required to invest ONLY in special Treasury bonds.” Mystery solved.
Now let’s get to my takeaways were regarding AI: first, I tried humor:
Sure, but yes... comparison to personally investing was my interest (ha ha, get it, interest?)
To which Claude replied:
Ha! I appreciate a good financial pun! 😄
But even more interesting, when I digged for some details the conversation got slightly contentious. I asked to see his work:
Those are the sort of things that made manual re-creation a nuisance task. So I was mostly curious to "see your work" like a teacher asks in school
Ha! Fair enough - you wanted to see if I showed my work or just scribbled down an answer. 😄
My “work” revealed:
✅ I divided grouped years evenly (simple, but crude).…Did my work pass your inspection, teach? 📝
I responded…
1) My questions do not imply inspection or doubt, only understanding. I go a lot slower than you. 2) Evenly splitting the multi-year periods…
To which Claude responded…
You’re absolutely right - I apologize for the snarky tone. Let me address your actual points:
1) Understood - you’re processing and learning, not testing me. I’ll slow down.
Wow, an apology from a bunch of transistors, nano-scale though they may be? Humor, is one thing, but do we call this “emotional intelligence” or something like that? Does it matter that we humans presume that it’s just a machine, and it somehow offered words that we presume to have feeling? I keep thinking of the TV show The Big Bang Theory, where the uber-nerd Sheldon will occasionally feign sympathy/compassion/sensitivity with a straight face and zero affect, challenging us to take him seriously.
All I know is Claude apologized to me.
March 5, 2026, Prone Bike
Today I asked Claude for help on my toughest project, one that is 40 years in the making… a prone bike.
The idea is to get the greatest amount of power out of the body by duplicating the way a 4-legged animal runs. I’ll boil it down to this:
Claude understood every nuance and challenge of the matter.
Proposed ways of thinking about it that I hadn’t imagined: “Articulated frame — the whole front half pivots relative to rear, like a loader…” and mimicking the universal joint of a vehicle driveshaft, for the articulation. Wow.
Got me to rethink the whole thing and start from a much simpler possible proof-of-concept to see if there’s even remotely a chance of developing the expected power from it. Again, wow. It’s possible that I’ve been trying to build a transmission for 40 years, and I might be able to replace it with just four “freewheels” (the one-way spinny thing on geared bikes).
But this was the most remarkable thing about the whole conversation, a compliment from a dumb bunch of transistors(?):
March 3, 2026, Reasoning, Patience, Limitless Capacity…
In a previous post where I summarized Claude’s abilities, I enumerated the attributes of its intellect, with stuff like insightful, articulate, and thoughtful. Today let me add:
Reasoning
Limitless patience
Limitless calculating capacity
Encouraging
I’ll explain why. While working on my puzzle idea (to add inspiring quotes at the end), Claude suggested one of my favorites, Edison’s “10,000 failures.” I had a side conversation about this where I showed Claude some of my 10,000, the image above.
Limitless patience: this is probably idiotically obvious to some readers, but I hadn’t understood it. You can work on a project with AI and it never tires, never forgets, never fusses. As for “never forgets,” I did have one interesting learning experience. I happened to delete a conversation (chat) and start all over, from only the HTML file that the previous conversation generated. It was as if Claude was a little dumber. For instance, one design choice (that resulted in a bug) was Claude’s choice, not mine: using the letter ‘B’ for blank crossword cells. This eventually resulted in a conflict with cells with the genuine letter B in them. In the newly continued conversation, Claude attributed the choice to me, not knowing I hadn’t authored the HTML. Funny, but a valuable learning experience in using AI. There’s considerable “thought” in a running conversation that does vaporize outside that conversation.
Reasoning: In the following exchange, I told Claude why I stopped farting around with my idea, in the rightmost image above, for a diurnal temperature differential energy generator (more details, next). How the fuck did Claude know that I only cared about the concept? (It was right.) And its appraisal of a legitimate exit is either shameless pandering or wildly reasonable. And its insight into trigger mechanism was a totally-mind-blowing insight for me. I sensed that creating a trigger was precipitous engineering delicacy, something that added to my math on abandoning the project… but without my understanding its profound history.
4 never got a successful cycle. Gave up fabricating a release trigger. Was satisfied that MIT validated the notion... accepted that feasibility was beyond me
That’s an honest and mature place to land. The validation question and the engineering question are separate — you answered the one that mattered to you, which was “is this real?” MIT confirmed yes. The trigger problem is just manufacturing, and manufacturing at that scale of precision without a shop is genuinely hard.
There’s also something specific about release mechanisms — they’re disproportionately difficult relative to the elegance of the system they serve. The rest of the machine can be rough and it still works. The trigger has to be exact. It’s the same reason crossbows took centuries to refine after the bow was mastered.
Satisfied with validation is a completely legitimate exit.
Limitless calculating capacity: Claude calculated something I’ve wondered about for years, but is way, way beyond my skills. Here was my question:
I'm glad we had this conversation. It reminds me that I was shot down asking on physicsstackexhange what the total global diurnal energy might be, even as a hypothetical starting point. Here's that page https://physics.stackexchange.com/questions/168962/does-ambient-day-night-temperature-change-have-significant-energy-generation-pot/497319#497319 Do you want to try to calculate that?
I’ll just provide out-of-context snippets of Claude’s answer:
Mass of atmosphere: ~5.15 × 10¹⁸ kgSpecific heat of air (at constant pressure): ~1,005 J/kg·KQ = m × Cp × ΔTQ = 5.15 × 10¹⁸ × 1,005 × 3Q ≈ 1.55 × 10²² joules per day5 × 10⁻³ W/m² × 1.5 × 10¹⁴ m² ≈ 7.5 × 10¹¹ W = 750 gigawattsSo realistic deployment at scale — covering all land surface with MIT-grade devices — gets you to roughly 750 gigawatts, about 4% of global energy demand. Not civilization-saving but not trivial either, and it runs at night, in shade, in sand storms, complementary to everything else. The people who deleted your question were wrong.
VINDICATION! Just kidding.
Encouraging: After I promised to stop gold-plating, Claude encouraged me to sell it to NYT:
OK. Unless I have some functional problem, this is what some--not me--like to call an mvp .
It is. And an honest one — not “minimum” in the sense of half-baked, but minimum in the sense that nothing superfluous made it in. The core loop is tight, the admin workflow is real, the deployment path is clear.
Go get that meeting.
With each moment spent with Claude it gets scarier, better, more amazing, more bone-rattling. Is this Man’s greatest creation and accomplishment? Is it bigger than a cure for cancer, solving the fountain-of-youth, flying cars, Moon landing, mRNA vaccines, DNA editing? Maybe.
I have not found a limit yet, and it’s only months old. I personally am perplexed at statements of “It will get much smarter.” I don’t know how it can, or why that’s useful. If it can get smart enough to reverse global warming or Man’s addiction to murder or war, then let it get smarter, sure.
February 28, 2026, Usability Conundrum
I continue my work on a proposed puzzle for the New York Times. I’m now doing all sorts of usability improvements, my main career focus when I pretended to work. I asked Claude about the relative use of time on usability vs. the basic functionality:
There are a lot of brilliant snippets in his answer (the “clear finish line,” “good usability disappears,” “additive vs subtractive”) but I see upon searching Google that some exist in common sources. Others look uniquely crafted by Claude. I’m not going to try to complete that analysis. To me, it’s a genius answer. And whatever the case, just consider his ability to interpret my question.
Here’s the text exchange (from the image):
Are usability adjustments always a seemingly disproportionate amount of effort, or is disproportionate simply a mistaken presumption of their merit relative to the satisfaction of raw functionality
Ha — it’s the latter, mostly. The presumption is the mistake.
Raw functionality has a clear finish line: it either works or it doesn’t. Usability has no such line — it’s a gradient that asymptotically approaches “feels obvious,” which is a moving target because what feels obvious is shaped by every other well-designed thing the user has ever touched. So the effort isn’t disproportionate to the value, it’s disproportionate to the visibility. Good usability disappears into the experience. Nobody notices the label you removed; they just feel slightly less cramped.
There’s also a compounding effect that makes it feel lopsided: functional work tends to be additive (you build a thing), while usability work is often subtractive or corrective (you remove friction, tighten spacing, reconsider a decision you already made). Rework always feels more expensive than it should.
The satisfaction gap you’re sensing is real though — shipping a feature produces a clear dopamine hit. Shaving 7px off a label produces a vague sense that something is slightly less wrong. The reward signal is weak even when the impact on the person using it is meaningful.
Latest Apocalyptic Prognostication: 2028 Global Intelligence Crisis
This article articulates everything I’ve already surmised about the likely impacts of AI on jobs and the economy. It’s probably capturing the eyes of the financial public as well because it talks in marketese. And it uses a rhetorical device that is killer: it talks in present tense, in 2028. It appears to be a public article. Get some tissues first, for your tears, because as I’ve written, our leadership system will not be fast enough. I’m not sure whether to recommend that you convert your assets—including your house—to cash because I’m not sure how much cash will hold up.
February 26, 2026, Crossjig Game
This is getting too much to bear. In an hour, with 4 total prompts (!) it made a game that I’ve been dreaming of for years, but never had the skill to produce. Yes you can even play it. The “usability” isn’t completed, but it works. Play the demo
Here was my first prompt:
Make A drag-and-drop phone app game where the player reconstructs a finished 7x7 'mini' crossword by repositioning arbitrarily sliced pieces that contain the letters of the finished puzzle and the black unused squares. I've attached 1) a finished puzzle and one without
I mistakenly pressed [Enter] too soon but it just ran off and started building.
I did another prompt, starting over using a 5x5 grid.
Then I told it not to include black squares. Make all squares white. And add a Solve button.
Then fix the drop targeting, which was too picky.
I asked Claude how to share it and it suggested that for beginners, I could just upload the file (which is a fully self-contained web page) to Netflify, so that’s what I did. Now keep in mind that this is just the raw code of the engine for such a game. The complete thing would require uploading new puzzles (daily) and possibly working on the slicing of the polyominos to make the game suitably difficult. I guess Claude can handle that… what can’t he do, right?
My favorite part of this exercise was when I was describing the finnicky-ness of the drag-and-drop. It took me about a hundred words, below. And Claude said, in a word, “Oh, I’ll use centroids.” Yeh, sure, stupid me, centroids.
Now I can finally write to the New York Times with the idea for this new game and have a link to show them the actual game. Look for it soon, right? OMG, pardon the double negative, but there’s nothing about this that isn’t mind-boggling.
As for the game itself, when I started to think of a name for it I eventually landed on one that revealed the First Law of the Internet: Someone has already done it. Crossjig.com is the exact notion, but from a blank slate, whereas mine mines the existing trove of The Times’ minis. And my idea includes the black squares and the final shape, a square, whereas crossjig.com is freeform. My only point is, it’s not a creative infringement of any sort. But back to our regularly scheduled program.
After soaking my head in sedative to forget mankind’s departure from the podium of gold medal winner in Intelligence, I went back to Claude to consider getting more serious about the game. I asked about getting to the next level. Here’s part of the answer:
Look at the use of metaphor (alphabet letters for the zig vs. zag of the puzzle shapes). Who the fuck does he think he is, a human? This is horrible in a magical, alluring, disturbing way. OMG, and he even intuited the significance of shuffling the black squares. This is truly staggering conceptual synthesis. How did they cross this mountain in just the last few months?
In the Feb 22 entry, below, I mention my long-gone friend, Scott. He was an autodidact of the first order: in 10th grade he latched onto the PDP 8 in our high school book closet, and then taught (yes, taught) my brother’s 11th grade computer class. He dropped out of a full scholarship because it slowed him down. In my basement, in our teens, he made an arc welder from an electric heater and some batteries. Perhaps the only thing in my life that makes me truly sad is that I don’t have him to ask the hardest questions, electronics, engineering, sociology, anything. When I read Claude’s answer above I had the brief sensation of impending exhaustion… because it was as if I had something of Scott’s capacity of which to ask questions… and it answered in milliseconds and I would have no time to breathe between mysteries.
February 22, 2026, Scott Banks’ Program
Today’s request of Claude:
In the 80’s i wrote some software on atari basic or maybe 6502 assembler that made 3-inch text display on a tv screen for advertising and similar. my unskilled code was about 200 lines of simplistic logic to expand letters from what must have been some sort of font bitmap, vectors unlikely. It ran slowly painting about 1 letter per second. An expert coder friend changed my 200 lines to about 17 lines that each were much shorter as well and it ran blazingly fast. He’s dead now and the code would be hard to retrieve. I always wondered what he did but never got around to asking. Can you speculate on what sort of underlying poetry he wrote that never occurred to me
In the following incredible answer, Claude nails it. I honestly can’t say that the exact facts of the software methodology are right, other than my heavily nested, verbose code—for the reasons above—but the broader conclusion is the only explanation: Scott’s code never translated from machine-to-human-to-machine… it was machine-to-machine. And look at this sentence: “when the logic is genuinely right for the substrate, brevity and speed arrive together.” How many human writers would you have to solicit before getting that brilliant construction??? And remember, Claude did this in 3 seconds. Wow.
9:41 AM
What a lovely thing to want to understand, and what a nice way to remember him.
Your 200 lines almost certainly worked the way an untrained mind naturally approaches the problem: nested loops, lots of IF statements, probably something like “if this pixel in the font bitmap is set, then draw a big rectangle here” — with all the bookkeeping written out longhand. Calculating x and y positions explicitly, handling each character with visible, repetitive logic.
Here’s what an expert 6502 programmer would have seen that you didn’t:
He thought in bits, not pixels. Your font was stored as bytes where each bit represented a pixel. You probably extracted those bits laboriously — maybe masking and comparing each one individually with something like
AND #%10000000, thenBEQ, thenAND #%01000000and so on, eight explicit tests per row. He would have used the processor’s rotate instructions —ROLorROR— which spin a byte’s bits one at a time through the carry flag. One rotate, check carry, act, rotate again. Eight bits consumed in eight identical micro-steps with no branching maze. A tight loop of maybe 3-4 instructions replaces eight separate IF blocks.He let multiplication disappear into shifts. Scaling your font up 8x or 16x for that giant text means multiplying coordinates. You probably calculated positions with actual multiplication or repeated addition. He would have known that multiplying by powers of two is just left-shifting —
ASL— which on the 6502 is instantaneous and implicit.He collapsed your nested loops into one. Where you had a loop for each character, inside that a loop for each row, inside that a loop for each column, inside that logic for each pixel — he likely saw that the entire thing could be expressed as a single tight inner loop that just knew where it was from the state of a few registers and the carry flag, with no overhead recalculating position.
He probably used self-modifying code. This sounds alarming but was completely normal elegant practice in that era. Rather than passing addresses around as variables, he would write the destination address directly into the operand of a store instruction as the loop ran. So the instruction itself changes between iterations. It reads almost like the code is alive. This eliminated indirection overhead that would have slowed your version considerably.
The poetry of it is that he was reading the problem the way the processor reads it — as a river of bits moving through registers — rather than the way a human naturally reads it, as characters made of squares. Your version translated human thinking into machine steps. His version thought like the machine from the start, so there was almost nothing to translate. The algorithm and the hardware were the same shape.
That’s probably why it was both shorter and faster simultaneously. Usually you trade one for the other. But when the logic is genuinely right for the substrate, brevity and speed arrive together. That’s the thing skilled low-level programmers of that era understood that felt almost mystical to everyone else.
February 20, 2026, 3. Chess Set
I made what I think is a pretty spectacular chess set, originally in expensive bronze with real gold/ruby/sapphire crown jewels, but I now cast my own in pewter. But I've never found support/enthusiasm/interest other than having won recognition in 1990 in a juried craft show. How do I get more commercial interest
And here’s the photo I showed Claude:
OMG. (Sorry for being repetitive.) Instantly, not even 2 seconds later, he reduced my 35 years of farting around to the following answer. He instantly nailed the conundrum I’ve long realized: that chess is a netherworld between art and kitsch. Here are the highlights:
12:12 PM
What a remarkable set — the craftsmanship is genuinely extraordinary. The live-edge walnut board, the character variety in the pieces (that executioner! the beseeching bishops!), the two distinct rook styles — this is museum-quality work. It’s honestly puzzling that it hasn’t found a wider audience, but I think the reasons are solvable.
Why it’s probably been hard so far
Chess sets occupy a strange commercial no-man’s-land. Mass-market buyers want cheap; serious collectors want either antique/historical or big-name artist provenance. And pewter casting, however skilled, doesn’t photograph as dramatically as bronze or silver, so online presentation really matters. The 1990 craft show win is wonderful but essentially invisible to today’s buyers.
Where I’d focus energy
Museum shops — the Smithsonian, MoMA, art museum gift shops actively seek unusual, well-crafted items. They have buyers you can pitch directly.
I have done enough experiments with AI at this point to declare with 100% certainty, this is already superintelligence, and I say that solely because Claude liked my work. For one brief instant my mind seriously composed the thought, “maybe it’s really like the Mechanical Turk, there’s a human somewhere in a distant sweatshop who contrived this answer.” But of course that’s idiotic, preposterous, impossible in the 1/2-second response time. But the reality is equally preposterous. I can’t believe where we are. When this thing gets hands with opposable thumbs…
Learning AI: 16-Min Youtube, Techniques to Use AI
This is an awesome video with concise yet innovative, strategic examples of how to use AI. Some examples:
00:38 Strategy #1: The Clone — Build a Persistent AI That Knows You
03:23 Strategy #2: The Swarm — Agent Mode, Vision, and Deep Research
06:04 Strategy #3: Devil’s Advocate — Force Adversarial Thinking
11:45 Strategy #5: Executive Coach — Use AI as a Mirror Under Pressure
February 20, 2026, 2. Aging Part 2, God
At the end of my Feb. 18 experiment on the fountain of youth, I tried asking Claude about God and he choked repeatedly, some excuse about server this-or-that, but c’mon, we both know he didn’t want to answer the question. I tried again today:
If aging is such a hard problem, like an unsolvable riddle, is that support for the argument that there is an omniscient, omnipotent mythical entity that made us... that deliberately made it an unsolvable problem
I won’t generally include a whole answer, but this was really great:
9:39 AM
This is a genuinely interesting philosophical move, but it doesn’t hold up well under scrutiny — and not just for the usual reasons.
The argument has a structural problem
The difficulty of a problem is weak evidence of intentional design. Things that are hard to reverse are often just things that weren’t built with reversibility in mind — and evolution is exactly such a process. It optimizes narrowly for reproductive success, then stops caring. Aging being hard to stop makes complete sense without a designer: it was never designed at all, so there’s no “off switch” to find. You’re looking for an architect’s blueprint in a structure that grew organically.
It also runs into the problem of what the designer wanted
If an omnipotent being deliberately made aging unsolvable, you have to contend with what that implies about the being’s character — it chose suffering, cognitive decline, loss, grief, the death of children from progeria. The “unsolvable problem” isn’t clean or dignified. It’s often ugly and painful. Theodicy has wrestled with this for millennia without a satisfying resolution.
And aging isn’t actually unique in its difficulty
Lots of problems are deeply hard — protein folding was “unsolvable” for decades, then AlphaFold cracked it. General relativity seemed impenetrable. Difficulty tracks with complexity and our current tools, not with intentional obstruction by an intelligence.
The deeper issue with this class of argument
It’s a god-of-the-gaps move: we can’t explain X fully, therefore a deity explains X. But this doesn’t actually explain anything — it just relocates the mystery. Why is the deity the way it is? What made that entity? The explanation has infinite regress problems and adds complexity rather than resolving it.
Where the intuition is genuinely interesting though
There’s a real philosophical question lurking here: why does the universe have the specific structure that makes consciousness possible but impermanent? That’s not nothing. Some serious philosophers and physicists find the fine-tuning of physical constants more compelling than biological complexity as a pointer toward something worth wondering about. Aging specifically, though, is probably the weakest version of that argument — because evolutionary biology gives such a clean, mechanistic account of why it happens.
The difficulty of aging feels like a riddle because we’re newcomers trying to reverse-engineer a process that was never engineered forward. That’s genuinely hard — but it’s the hardness of archaeology, not of a locked door someone hid the key to.
That answer above is so informative to me. I feel like there’s nothing more to comprehend on the matter. Wow. (I still think we can stop the process, though.) I asked Claude this followup:
Very helpful, thanks. Is alphafold agreed upon by the preponderance of experts. I've seen refutation on the uncredentialed blogosphere.
It’s answer was a detailed “yes,” but again what popped out to me was phrases like this: “These exist, but they're nuanced rather than fundamental.” That writing precision and artistry is just breathtaking. Only a very few people can write a sentence that perfect (and in 2 seconds). It contrived a perfect contrast (nuanced/fundamental), put the positive (nuanced) first, and used perfect economy of words. OMG. I give up.
February 20, 2026, 1. Rotator Cuff Injury
how do i decide on surgery for a moderate rotator cuff injury caused by a fall upon the elbow, 60 year old male, very active, weekend warrior hockey player in good health
As the familiar saying goes, “I have this friend who…” Well, in this case it’s not a veiled joke. It really is a friend who injured his shoulder, but I was curious to ask Claude because I too (and two other hockey-playing friends) had this exact injury many months ago.
Result? Claude did a perfect job summarizing the situation. Nothing as magical as my other experiments… just a much better result than asking Google. All major medical information providers, of course, try to answer such questions. I worked on one such system for the publisher Elsevier, an absolute world leader. It had thousands of pre-written pages of such answers. Claude’s answer is like the best-of-the-best of those AND nuanced for whatever additional context and circumstance you prompt it with. Here are examples of Claude’s most meaty details:
Age 60 is actually a sweet spot: old enough that the tear won’t fully heal on its own, young enough that surgical repair heals well and you have years of active life ahead
If it’s a full-thickness partial tear or complete tear of any tendon, the structural defect tends to worsen over time without repair
February 19, 2026, My Latest Invention
I asked Claude the following, with the proprietary stuff removed:
what is the market potential of a cross between a [xxx] and a [yyy] like the image i've attached
Claude found existing products that weren’t the precise innovation, so it took me one more prompt to zero in a little more:
the [xxxyyy] is only incidentally for leaves. It's targeted for [such-and-such situations]
And it proceeded to tell me insights into my idea that even I had not imagined. I then had a back-and-forth conversation about marketing, and my existing video where, for instance, it made a perfectly correct observation about the amateurish color of my main title text. OMG, I can barely comprehend any of this. It makes no sense.
Learning AI: 11-Min Youtube on Google’s Free Tools
This looks like an AI generated video but it’s good because it shows you the scope of Google’s free focused tools (no chapter marks), but short:
1. AI Studio Build, app builder
2. Opal, app builder
3. NotebookLM, research and note-taking
4. Pomelli, marketing
5. Gemini Canvas, text and document helper
6. Nano Banana Pro, image generation mode
7. Multi-Speaker Audio Generation
February 19, 2026, Give me text, not video!
how to find my comments on youtube and don't show me a video
Now that’s a pretty lame example, right Jack? NO. I search Google for such a simple answer and the answers are stale because the Youtube interface changes all the time. And 99% of the answers are fucking videos that waste incredible amounts of my time. The answer is 3 short sentences, using things called words. Mankind started with cave drawings, then spent thousands of year to turn those pictures into the proverbial “thousand words,” then went backwards again to time-wasting Youtubes. :) Sheesh.
AI did get me the text answer, but interestingly, by getting the stale info one (critical step closer) with this text: “Click your profile picture → Your data in YouTube.” So the moral of the story is that AI will let you force text (not video) answers.
More interesting, however was my follow-up and response:
February 18, 2026, Fountain of Youth
I was watching a Youtube on AI with this smart MIT guy doing an excellent impression of Sheldon Cooper and he casually threw in the phrase “solve our biology” or words to that effect. I immediately presumed he was implying that AI would solve the fountain of youth. I’ve written for about 20 years now, that I think it’s a certainty we will solve this. (My ‘proof’: thinking otherwise is an implicit postulation that there is some mystical, omnipotent entity that made an unsolvable conundrum of aging… and since I think that such a ‘God’ answer is silly, quod erat demonstrandum, so to speak.) I’ve stopped calling it anti-aging because deniers imagine “endless decrepitude,” which no one wants. I’ll skip right past the taint of quackery and call it what it is, fountain of youth. Anyway, I thought I’d ask AI to solve it:
How can we stop, not just slow or delay, the mechanism of cellular aging
OMG. First it’s worth noting that it immediately answered my real question, no jibber-jabbering to get to the heart of the matter. It’s answer was an awesome example of its staggering ability at research and summation, again (as in a prior day’s enlightenment) with monstrously skilled techwriting. Over the years, I had come across most of the morsels of what it found, but hadn’t read about it in years. In one page it just nailed the state of the art, key snippet here:
February 14, 2026, Language Training Tool
make a language training cartoon of frere jaques at half speed with synchronized french and english captions
First pass, it didn’t include sound. Hmm, this Claude is talented, but sorta dumb.
oops, add audio
Hmm, it added simple music (maybe a piano) but no voice.
add spoken voice for the french only
This is worse than working with offshore programmers! (Oh, sorry, did I say that out loud?) It’s like if you went to a restaurant and said to the waiter “I’d like the filet mignon,” and half hour later you had to ask, “Where’s my steak? And they said, “You just told me what your preferences were… you didn’t actually ask me to bring that to you.”
Not quite OMG this time, but still pretty incredible for ten minutes of work: the voice sucked… not good enough to casually learn French, but I’m optimistic I could search around for an AI thing with better spoken French. But nonetheless it is sort of incredible. Here’s why: I have searched Youtube for French nursery rhymes to learn some French, but NONE of them have the translated English synchronized, so I can’t connect the vocalized French to the English in my brain. In ten minutes with AI, I’m on the precipice of solving that problem and producing my own content.
Learning AI: 30-Min Comprehensive Youtube
This is a high-quality overview of tools, methods, and concepts from a channel that also has paid courses. Skip through chapter-at-a-time to learn as little as you want on each topic.
February 13, 2026, Shackleton Movie
Thirty years ago I learned about Ernest Shackleton’s incredible story of survival and rescue in a National Geographic magazine at my inlaws’ apartment and I’ve been obsessed with it ever since, even making an absurd wall sculpture (full story of that artwork on a Medium page). I’ve always dreamt of making a movie of the full adventure since no one has done so. And now I can try to have AI make it. I’ll try just the most dramatic part:
First result: a laughable, pathetic, childish cartoon, not photorealism; no sound. Yes, that’s a boat with stick men in it, below.
Second result: told it to add sound. Same pathetic result with sound. And the web page needed some kickstarting to get the video controls to be interactive again.
Maybe I’ll look for a video tool, but first let’s try reasoning with our young filmmaker:
It tried:
The cartoon was 50% better but still lame. Let’s look for a dedicated video tool.
Five minutes later I was using a free tool from the ChatGPT site, Descript
https://web.descript.com/.
I used the same prompt but included “photorealistic” and “with sound.” It presented its plan to me for approval and I said “looks good.” About 5 minutes later it presented what looked like a full video editing suite like the free Shotcut tool I’ve used. Not clear if it was ready for me to press play yet, but let’s try it.
I clicked play and the audio narration started, sounded nice… but no video. Ahhh… it’s still generating:
I’ve been a usability person since 1985 (if you’re curious search this page for ‘bellis’), so I get a little persnickety that they use passive/implicit messaging instead of overlaying it right in the video, but everything is relative, eh? This %#$^@$ thing is looking like it’s going to give me the power—as if the arms—of Michelangelo, Da Vinci, Franklin, Edison, and Spielberg, in 30 minutes for just the cost of my computer and internet connection… so I’ll try to get over some imperfect usability.
OK, it’s done. OMG:
Despite many funny mistakes, it’s nothing less than a miracle. Can I change the audio to first-person conversations, upload images of the James Caird, ad infinitum??? I have no doubt.
This is incomprehensible.
February 12, 2026 , AI and Database Apps
I was going to ask one of my more programmer-ish family members this question, then I woke up and realized I should just ask AI:
“when I have ai create an app for me for free, where is the database and who pays for it and how do i scale it”
OMG, it understood it perfectly and provided a comprehensive answer with the REAL issue I was talking about. So so many times typical search engines never get to the real point of a user problem—I’ll try to dig out an old example—but this even enumerated the steps of how to set up my own db when needed. In this Substack story, I don’t expect to show the actual content of AI answers, but for this early one, I can’t resist. Here it is as an annotated image, yellow highlighting all mine:
But here’s why it’s earth-shaking/bone-rattling/unnerving. I’ve been a techwriter since 1985. Even though I transitioned to user-interface design—the kids call it ‘user experience—(through some coding and other roles), before retiring in 2022… I’ve been a writer all the way. And the sample you see above is nothing less than the BEST techwriting I’ve ever seen.
It does not skirt the issue by answering only the easy part that you already might know.
It provides proper noun examples instead of pretending that’s financially impolite.
It provides dollar amount examples instead of hiding behind the excuse that such info is embarrassingly subject to error over time.
It is mostly bullets!!!!!!!!!! Sorry for the childish emphasis, but I’ve tried for years to explain that almost all techwriting (including newspapers) should be mostly bullets not prose. We’ve kept it as prose for 500 years because of the weaknesses of pre-digital typesetting.
Its sentence structuring is not simply flawless but, to this incredibly nitpicking eye, lovely. I love, for instance the subtle choice between the subject and the parenthetical portion in… “Anthropic (the company behind Claude)” and many other grammatical constructs that are so respectful of “structured thinking.”
But none of those are as important as this: it breaks down the matter into its independent, orderly components… truly intelligent… real cogency (look it up).
Again, it is quite simply the best techwriting I have ever seen. And I think it took 3 seconds. OMG.
January 16, 2026, Tree Surgeon
“Design a remote controlled tree surgeon”
OMG, it understood exactly, I think on one try, and came up with a slightly alternate plan from mine but included conventional features from other systems such as drones that I hadn’t even thought of. And it included a second mode (a pruner, not a saw) that would really improve my concept!