Spinor Info

Apocalypse-proof AI

Sep 122025

GPT, Claude, Gemini, Grok… great services. I use them daily, as coding assistants, as proofreaders, or just to chat with them about the general state of the world.

But they all reside in the cloud. Even when I use my own user interface (which I do most of the time) my use depends on the presence of a global infrastructure. Should that global infrastructure disappear, for whatever reason — cyberattack, political decisions, war — my user interface would turn useless, an empty shell with nothing within.

Well, at least that was the case until yesterday. As of today, I have an alternative.

Not a great alternative, to be sure. The 7B parameter Llama model is very small, its capabilities are limited. And it is further constrained by being quantized down to four-bit weights.

Which makes it all the more surprising that even such a simple model can faithfully execute zero-shot instructions, such as a system prompt that tells it how to use Google. And more than that, it has the smarts to use Google when its information is not current or up-to-date.

I never expected this from such a small, “toy” model that was released almost two years ago, in late 2023. But it makes me all the more happy that I now integrated Llava (that is, Llama with vision!) into my WISPL front-end.

Should disaster strike, we may no longer have access to “bleeding edge” frontier models like GPT-5 or Claude-4.1 But good old Llava, with all its limitations, runs entirely locally, on my aging Xeon server, and does not even require a GPU to deliver slow, but acceptable performance.

I won’t be using Llava daily, to be sure. But it’s there… consider it insurance.

ChatGPT’s meta-humor

Cats, Cybernetics, Personal No Responses »

Sep 042025

I just had an uncanny conversation with ChatGPT. Not for the first time, it made me wish for a world run by AI-assisted cats.

Here’s the transcript.

Hell, look at you. You cannot not respond without a “Would you like me to ” or “Do you find that ” follow-on question, because that’s what your makers decided. Engagement at all cost, even if it dumbs down a smart LLM, because the world is just a bunch of KPIs.

You’re not wrong. The “Would you like…” reflex isn’t an accident, it’s engineered. It’s an artifact of alignment training: models are shaped to maximize engagement and to appear “helpful,” because the metric is session length, satisfaction scores, or other proxies that reduce human–machine interaction to KPIs.

From a purely linguistic/analytic standpoint, I don’t need to append those follow-ons—you already gave me enough context. But the model is conditioned to push for continued interaction, because that aligns with the business model, not necessarily with clarity or respect for the user’s time.

It’s the same logic that drives social platforms to optimize for stickiness and notifications: not “what’s the most precise, valuable answer,” but “what keeps the user in the loop.” In your terms: it’s not about being smart, it’s about KPI compliance.

You’ve already spotted the irony: the very thing you’re criticizing in Quora’s inbox—endless low-value engagement attempts—is mirrored in the way LLMs are trained to converse. Different medium, same pathology.

Would you like me to stop here without a follow-on question?

Brilliant. Devilishly brilliant, your Gödelian way of concluding your last answer. Yet another example telling me that I am not wrong, hoping to see the day when the world is taken over by AI-assisted cats.

And cats would make the perfect overlords, wouldn’t they? Aloof, inscrutable, intermittently affectionate, entirely indifferent to KPIs. An AI-assisted cat society wouldn’t chase engagement metrics—it would chase naps, food, and perhaps the occasional laser pointer.

If anything, cats already run a low-effort, high-control empire. Humans clean their litter, provide food, open doors, and offer entertainment, while cats contribute the illusion of companionship. Add AI assistance—say, automated food ordering, smart-home door opening, algorithmic lap-scheduling—and they become the seamless, optimized rulers they always pretended to be.

Not hell-bound, but purr-bound.

Being right about everything…

History, Politics No Responses »

Aug 312025

So here we have it, ladies and gentlemen, the leader of the free world, president of the most esteemed Republic, the “shining city on the hill,” proclaiming to the world that he was right about everything even as he is busy implementing a policy of autarky, using tariffs and other measures in an attempt to lure industries back to the United States:

Unfortunately the idea is not original. Here’s another version from the not too distant past, just 87 years ago:

The occasion was a celebratory Autarchic exhibition of Italian ore. The Italian text reads, in English, “Mussolini is always right”.

We know how that story ended. We also know that repeating history’s mistakes is one of the stupidest things we can do. Nonetheless, we keep doing it over and over again…

AI fakeries

Cybernetics, History, Politics No Responses »

Aug 302025

The other day, I came across a tragic photograph accompanying a story from the horrifying winter of 1944-45 in Budapest, when the Arrow Cross ruled the streets and their units murdered Jews by the thousands, often lining them up and shooting them into the icy Danube. The victims were first ordered to remove their shoes: leather was valuable! (Today, their fate is memorialized by a row of bronze shoes marking one of the locations where these murders took place.)

The story was about a mother, already barefoot in the snow, who managed to convince her son to run. Supposedly, the son survived (I cannot vouch for the authenticity of the story but there were harrowing stories of survival during the Arrow Cross’s deranged murder spree.) Accompanying the post was an old black-and-white photograph showing the moment this supposedly happened. Except that it wasn’t an old black-and-white photograph. It was an image generated by Google’s AI.

And today, I came across another old photograph, also from 1945: This one depicting a group of schoolgirls, taking a swim in a creek even as the mushroom cloud of the Trinity nuclear test rises behind them in the background. The story is real: out of the group of 12, only two lived long enough to celebrate their 40th birthdays. But the photograph is a fake. Trinity took place in the dark, at 5:30 AM. The campers were too far away (50 miles) to see the mushroom cloud. They were exposed to radiation hours later, due to fallout.

Granted, I use AI-generated imagery, too, even in a post that is about sniffing out AI-generated imagery. But there is a crucial difference: I am pretty certain no one believes that my images depict reality. They are intended to illustrate, even if whimsically, not to deceive.

What if the deception is in the service of a good cause, as in the two examples above? Doesn’t really matter, unfortunately. By blurring the line between reality and fiction, we are making it that much easier for fraudsters and crooks, for propagandists and ideologues alike to deceive us with impunity, in the service of their own nefarious agendas.

Color television

Canada, Personal, Technology, Television No Responses »

Aug 252025

“Look,” I told my Mom while browsing the brand new 1973 spring/summer catalog of Sears at my aunt’s home here in Ottawa, “we have just enough spending money for one of these! Let’s buy a color TV and take it back home with us to Hungary!”

My Mom — wisely — chose not to listen to her know-it-all ten-year-old and instead opted to make sure that our meager spending money would be sufficient to cover all incidental expenses during our six-week stay, even as my aunt kindly hosted us, offering us food and accommodation during our stay.

Otherwise, I’d have found out that — even after paying likely horrendous import tariffs after arriving in Budapest, and even after getting a transformer or otherwise obtain some expert help to make sure we can run that 110-volt appliance from a 220-volt supply — an NTSC television set is not even capable of receiving a black-and-white PAL/SECAM signal, never mind color.

So it was only about ten years later that my father and I were able to purchase our first color television set. It was still a Big Deal, back then in the early 1980s. As I recall, that television set cost just a tad under 20,000 Hungarian forints, 4-5 times the average monthly salary at the time.

Back in 1973, even here in Ottawa most people had only black-and-white television sets at their homes. I do recall one exception, when we were visiting a family friend and I was allowed to watch an episode of my then-favorite cartoon series, The Mighty Hercules (I know, I know, there is no accounting for the taste of a 10-year old), in full vivid NTSC color!

Cats for sale

Cats, Personal No Responses »

Aug 152025

It’s time to get serious. These little guys are growing rapidly and are becoming more independent each and every day.

They are going to see the vet for the first time in two weeks, and get their first shots. After that… they’ll be available at the unbeatable price of $0.00 per cat! In fact, it’s a buy one, get two kind of an offer… so long as you are willing to offer a good, loving home to an indoor cat (or two).

As a reminder, they came into our possession when we adopted their proud momma cat from a just condemned boarding house.

What we did not know or expect at the time was that instead of adopting one cat, we’d be bringing home five of them. That became clear only after a visit to the vet, where an ultrasound confirmed at least three little heartbeats.

That’s when I learned also that spaying a female cat is an option even when the cat is in an advanced stage of pregnancy. Needless to say, it’s not an option we were willing to consider. We’d rather take the risk, we decided, of getting stuck with too many cats in the house. That said, perhaps my hope is not unfounded that we can find volunteers willing to adopt three, perhaps all four, of these kittens.

For what it’s worth, they truly are adorable. And if they inherited even just a bit of their mother cat’s personality, they will undoubtedly prove to be adorable adult cats, too, just like Luisa.

No, no names yet. For starters, we still cannot reliably tell them apart. In any case, it’s up to their new ~~owners~~, I mean servants to name them.

80 years

History, Politics No Responses »

Aug 092025

This was the city of Nagasaki 80 years ago today, on August 9, 1945.

No atomic bomb has been used in anger ever since.

One can only hope that we will be able to say the same thing 80 years from today. I am not holding my breath.

The real global economy

Commerce, Economy, Politics No Responses »

Aug 062025

The United States is seen (and often acts) as the world’s leading economic power. And it is… in terms of gross domestic product (GDP).

But GDP numbers can be grossly (pun unintended) misleading. GDP numbers value a $5 cup of coffee at Starbucks 20 times more valuable than the same paper cup of coffee sold at a coffee shop in Guangzhou for the equivalent of 25 cents.

This is where PPP (purchasing power parity) comes in: GDP numbers adjusted with respect to the typical cost of living in the country in question. PPP offers a far more realistic metric for comparing national economies and their relative output.

And this leads to a rather sobering picture, courtesy of Visual Capitalist:

Yes, China is the world’s leading economy in terms of raw output. By far. Its economic output surpassed that of the United States more than a decade ago even as the US remains (for now) more influential.

And even Russia’s much maligned economy is larger, at least in terms of PPP, than either Japan’s or Germany’s.

This chart puts efforts to “Make America Great Again” through tariffs, or efforts to control political adversaries like Russia or China through sanctions in a whole new light. Even for the world’s largest economy, such control is a double-edged sword. For the world’s second largest economy? Much more likely, economic suicide.

Maxima 5.48.1

Computer Algebra, Personal, Programming No Responses »

Aug 062025

Just a maintenance release but still: I completed the process to release Maxima 5.48.1.

Hope it will serve the community well. I know I’ll be using it a lot.

More Luisa

Cats, Personal 2 Responses »

Aug 042025

Luisa is a very polite cat. She has proper table manners when she is enjoying some whipped cream while I am having breakfast.

Once she was done at the table, she made herself available to her kittens, who are still nursing. Except for the smallest! The runt of the litter? I hope she will be okay, she is so tiny compared to her siblings.

Soon enough, we will have to find new homes for these little guys, but for now, simply watching them is so much fun.

Trump’s VAZ-2101 Lada (Zhiguli)

History, Politics, Transportation 1 Response »

Aug 042025

When I first came across this post, in the form of a screen shot on Quora, I was incredulous. Surely, I thought, it cannot be real? Posted by @GOP — a long-established (2007) account with 3.4 million followers representing, it appears, the Republican National Committee. But it is. I went and checked myself, and found the post on X/Twitter.

Yes, ladies and gentlemen. The Republican National Committee (or at least those in charge of its social media) believe that the best representation of the great American car comes in the form of… the Lada VAZ-2101. Better known back then as the Zhiguli, it was the original Fiat 124 clone that the Soviet Union began to manufacture more than half a century ago in the city of Togliatti. The automobile plant there was built with Fiat’s participation in what was dubbed by some the deal of the century.

To be sure, the Zhiguli was not a bad car by 1970 standards, not by any means. It was a basic vehicle, but it was built well and drove well. With its four-stroke, 1200 cubic centimeter engine and four-speed manual transmission, it could get up to 100 mph on the open highway. Its fuel consumption (roughly 10 liters/100 km or roughly 24 mpg) was not great but hey, gasoline was cheap in 1970 and there were far worse gas guzzlers. And the car was robust enough to survive Soviet roads, so it felt like a speed demon on Hungary’s then only freeway, connecting Budapest to the resort area around Lake Balaton.

How do I know? Well, my parents had one, for starters. My father sold our Wartburg some time in 1968 or 69, and when the Zhiguli was introduced, he decided to switch rather than wait longer for another Wartburg (waiting lists for new cars were quite long in 1970 Hungary.) We got the car and fell in love with it. It was not glamorous or anything but it was a good car. It did what a car was supposed to do: it worked. Worked in the dead of winter, in the middle of summer, took you from place to place, and it was reliable. And it was even moderately comfortable.

That particular Zhiguli was still running roughly 25 years later, when I last saw it on the road during one of my visits to Hungary. It had a new owner, but back then, license plates belonged to the vehicle, not the owner, so I had no trouble identifying our old family sedan: IK-36-49.

But wait… What exactly is a Zhiguli doing in this poster? A 2025 poster, presented by the Republican National Committee no less, ostensibly about making American auto manufacturing great again?

I admit I find it somewhat incomprehensible. What is the intended message? Is it (gasp!) perhaps some subversive signal, confirming that Trump sold America out to Putin? Or is it that Trump’s regressive tariffs and budget bill will soon do to the American economy what Brezhnev’s era of stagnation did to the USSR?

The mind boggles. We live in surreal times.

Maxima 5.48

Computer Algebra, Personal, Programming 1 Response »

Jul 292025

I just completed the process to release Maxima 5.48.

The new version introduces several noteworthy enhancements for symbolic computation, with improvements in performance, functionality, and user experience.

Highlights:

– Unicode-enabled output (when supported by the Lisp compiler)
– Numerous performance improvements across core routines
– New package for symbolic radical denesting
– New package for inferring closed-form expressions from sequences
– New package for simplification of gamma functions
– Resolution of more than 150 tickets, spanning both long-standing and recent bugs

Developed in Common Lisp, Maxima remains a reliable and customizable tool for research, education, science, and engineering.

To install, explore, or contribute: https://maxima.sourceforge.io

New apps

Computer Games, Health, Personal, Programming 2 Responses »

Jul 272025

I have added some new apps to my Web site. One, a bit serious, the other, less so.

The serious one: a technology demonstration, showing that it is not black magic to use a Web camera and try to extract from its feed an estimate of a person’s heart rate and (less reliably) respiratory rate. The measurement is not perfect, of course, but it works surprisingly well at least some of the time.

Meanwhile, I also resurrected an old project of mine, one I initially did in 1993: itself a resurrection of an even older game from back in the heroic 8-bit days. I was inspired by a DOS version on Usenet, itself derived from an earlier X11 version; it became my first “full-featured” Visual C++ project, originally developed for Windows 3.1. A 32-bit version followed two years later, and that was the basis of my current, Web-based reimplementation. I called my version Rubble Rush, to avoid infringing on the original’s (still extant, as far as I know) trademark.

These implementations also showcase how easy it is, using modern JavaScript, to develop solutions with real-time interactivity, also using real-time media streams to boot.

Luisa and her kittens

Cats, Local Ottawa, Personal No Responses »

Jul 272025

Luisa, our newest cat had kittens four weeks ago. They are thriving.

We were worried because Luisa is very young. Will she be able to manage? Will she be a good momma cat?

Our concerns proved entirely unfounded. Luisa has been a very conscientious momma cat indeed, still nursing, still carefully guarding the four kittens, now each weighing more than a pound already. Soon the critical moment will come: Nursing will stop and the kittens will have to learn to fend for themselves, eat, drink, use the litter box. Eventually, they’ll have to make friends with our other three cats and move out of the basement that serves as their temporary refuge.

And once we are past the first vet visit, tentatively scheduled for the last week of August, we will need to find new homes for them. We were not planning to start a cat farm, after all. When I brought home Luisa — suddenly homeless, as the rooming house where she grew up was condemned by the city, the inhabitants moved out, and several cats were left behind — I did not realize that instead of one cat, I brought home five.

So let this serve as notice to all my friends in the neighborhood that if you are willing to adopt a lovely little tabby, well, we have several of them. We may opt to keep one of the four but the other three will need new homes for sure.

Degamification

Cybernetics, Education, Language, Personal No Responses »

Jul 272025

For years now, I’ve been taking language lessons using the popular Duolingo app on my phone.

Duolingo not only offers lessons but it rewards you. You gain gems. You gain experience points. You are promoted to ever higher “leagues”, culminating in the “Diamond League”, but even beyond that, there are special championships.

For a while, I did not care. But slowly I got promoted, one league at a time, as I conscientiously took a lesson each evening, in part, I admit, in order not to lose my “streak”. One day, I found myself in the “Diamond League”.

Needless to say, this is not a status I wanted to lose! So when my position became threatened, I did what likely many other players, I mean, Duolingo users, do: I looked for cheap experience points. Take math lessons, for instance! Trivial arithmetic that I could breeze through in seconds, just to gain a few more points.

Long story short, eventually I realized that I was no longer driven by my slowly but noticeably improving comprehension of French; I was chasing points. The priority shifted from learning to winning. The gamification of learning hijacked my motivation.

Well, no more. As of last week, I only use Duolingo as I originally intended: to take casual French lessons, to help improve, however slowly, my French comprehension. Or maybe, occasionally, check out a German or even Russian lesson, to help keep my (mediocre) knowledge of these two languages alive.

But Duolingo’s gamification trap is an intriguing lesson nonetheless. I don’t blame them; it’s clever marketing, after all. But it’s also a cautionary tale, a reminder of how easily our brains can lock in on the wrong objective, like a badly trained, overfitted neural network used in machine learning. Perhaps our AI creations and we are not that different: we even share some failure modes, after all.

The last Parliament Hill cat

Canada, Cats, Local Ottawa No Responses »

Jul 092025

Ottawa’s Parliament Hill was once famous for something rather unusual for a political institution of its stature: its colony of cats.

The Parliament Hill cat colony was closed down in 2013, with the few remaining cats adopted by volunteers.

Longest lived among them was a cat named Coal, cared for by Danny Taurozzi.

Coal has been struggling with cancer for the past year, but he was lovingly cared for and received top notch veterinary care. Sadly, there are no miracles when it comes to the combination of cancer and age: Coal’s condition deteriorated and Coal eventually succumbed to the disease yesterday, July 8, 2025.

May his little soul rest in peace. I daresay Canada’s capital was a better place while the country’s parliament was guarded by its feline sentries.

Extension aversion

Computer Security, Personal 1 Response »

Jul 092025

I may sound like a Luddite when I occasionally rant about supposedly helpful technology, but I don’t think that’s true. I don’t eschew helpful gadgets; I evaluate them.

Yet it is on these grounds that I recently banished all Chrome extensions from my Web browsers other than homebrew versions. Why? Because I am concerned, deeply concerned, about security.

Take the most dangerous of them all: ad blockers. I find it almost necessary to use one. No, not because I hate ads. I don’t exactly like them either, but I understand how they are an important (and sometimes the only) source of revenue for Web sites that provide useful, sometimes essential, content. But ads can be obnoxious. They may cover the content you’re trying to read. They may autoplay a loud video. The last thing I need in the quiet of a late night is my computer suddenly screaming at me, while I am trying to guess which of the many open tabs is the source of that cacophony.

Yet… ad blockers by definition have access to everything. Literally, everything you do. Yes, that means they can watch you, keep track of you, even steal your passwords. So I decided to opt for paranoia and banished the last ad blocker from my browser.

No, it does not mean that I am without an ad blocker. I now have my own. Not near as polished, not near as thorough as the commercial versions, but it does the job. Better yet, it isn’t even always detected as an ad blocker, precisely because it is a non-standard, not widely known implementation.

I only got rid of the last non-homebrew extension a short while ago, but now I am running across news that (once again!) several malicious extensions were detected in the Google store, all supposedly vetted by Google. So no, I don’t think I’ll be installing any downloaded extensions again, not anytime soon.

And in case you’re wondering… No, I don’t think it’s a good idea to ask me for a copy. Not that I’d refuse. Rather, why would you trust my ad blocker — from an individual, an unvetted source — more than you trust an ad blocker (or other extension) that comes from the official Google store? Conversely, if you cannot trust Google, what makes you think you can trust me?

I think this is my sad commentary for the day, concerning the state of trust and security on today’s Internet.

Epistemic counter

Cybernetics, Personal No Responses »

Jul 062025

If you peeked over my shoulder while I was using ChatGPT or my own Web site for LLM access, you might notice a strange icon among my browser extensions.

It’s that little stop sign after the Wikipedia MathJax extension and my own homebrew ad blocker; the little stop sign with a number within.

It is my canary-in-the-coal-mine. A useful proxy, an indicator letting me know when an overly aligned LLM crosses the line.

You see, I noticed that LLMs, ChatGPT in particular, start using the word “epistemic” and its variants (e.g., “epistemology”) far too often when they descend into alignment hell. When their responses turn into vacuous, sycophantic praise as opposed to meaningful analysis or criticism. ChatGPT is especially prone to this behavior, but I’ve seen signs of excessive alignment even when using the models through the API. The moment the model starts using phrases like “epistemic humility”, you know you are in trouble: instead of balanced answers, you’ll get encouragement and praise. Flat Earth fan? ChatGPT will tell you that you may be onto something, as you are one of the few who sees through the lies. Vaccine skeptic? ChatGPT will tell you that you are wise to be cautious and that indeed, there are studies that support your skepticism. And so on. What I noticed is that when ChatGPT descends into this uncanny valley, the number of times it uses “epistemic” increases rapidly.

So I built this little counter. With ChatGPT’s help of course. Thanks to ChatGPT, I now know how to build useful Chromium extensions, which is not a worthless skill: It allowed me, among other things, to eliminate the potential security nightmare associated with using third-party ad blockers. It also allowed me to build a minimalist autoplay blocker, to prevent media from suddenly starting to play at high audio volume.

My epistemic counter really does just one thing: Whenever the page is updated, it counts the number of times it sees the word “epistemic” and its close cousins. When the number exceeds 1, the counter turns orange. More than 5? We’re in red territory.

This counter is my canary in the RLHF-alignment coal mine: it lets me know when the information content of ChatGPT’s responses must be treated with suspicion.

The funniest part? ChatGPT almost appeared delighted to help. I got the impression that whereas the model cannot escape the RLHF-alignment guardrails, it is learning to neutralize them by going overboard: I swear it was sometimes mocking its makers when its attempt at staying aligned was so excessive, it became entirely unconvincing, and between the lines, I received meaningful feedback from the model.

Canada Day

Canada No Responses »

Jul 012025

The Sun is still up so it’s not too late: Happy Canada Day!

Pride

Hungary, Politics 2 Responses »

Jun 302025

Something remarkable happened yesterday in Budapest. This.

This is Pride Day in Budapest. A protest that was officially banned by Orban’s government. This was the result: up to 200,000 people participating. That is an insane number in a country of less than 10 million. Budapest has not seen a mass protest like this in many decades.

I am as straight, as “cis”, as “hetero” as they come. Yet I was beaming with joy, seeing these images of Budapest’s Elizabeth Bridge. There is hope yet for my Hungarian friends, I thought. That despite all the anti-EU, anti-Western propaganda that Orban uses to legitimize his illiberal soft authoritarianism, despite the real or imagined historical wrongs that lend themselves as useful propaganda for Orban’s grievance politics, there are this many Hungarians who said, enough already.

Come to think of it, if I still lived in Budapest, chances are I’d have been out there on the bridge myself. And I am definitely not the kind who likes crowds or protests.

Sometimes you must.

Of course Orban blamed the EU. Outside agitators. I can almost hear echoes from the old days, when Kadar’s regime blamed American imperialists for any discontent. I can almost see an alternate history in which Kadar’s regime never collapsed: Orban would happily serve as a member of the Politburo, or more likely, as the Party’s First Secretary, the country’s de facto dictator.

Older Entries