<?xml version="1.0" encoding="UTF-8" standalone="no"?><rss xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:itunes="http://www.itunes.com/dtds/podcast-1.0.dtd" xmlns:media="http://search.yahoo.com/mrss/" xmlns:slash="http://purl.org/rss/1.0/modules/slash/" xmlns:sy="http://purl.org/rss/1.0/modules/syndication/" xmlns:wfw="http://wellformedweb.org/CommentAPI/" version="2.0">

<channel>
	<title>Il Blog di Marco Camisani Calzolari</title>
	<atom:link href="https://www.camisanicalzolari.com/feed/" rel="self" type="application/rss+xml"/>
	<link>https://www.camisanicalzolari.com</link>
	<description>Il Blog di Marco Camisani Calzolari</description>
	<lastBuildDate>Thu, 17 Sep 2026 15:01:22 +0000</lastBuildDate>
	<language>en-US</language>
	<sy:updatePeriod>
	hourly	</sy:updatePeriod>
	<sy:updateFrequency>
	1	</sy:updateFrequency>
	<generator>https://wordpress.org/?v=7.1.1</generator>
<site xmlns="com-wordpress:feed-additions:1">47191908</site>	<itunes:explicit>no</itunes:explicit><itunes:subtitle>Strategic Advisor | AI, Cybersecurity &amp; Digital Transformation</itunes:subtitle><item>
		<title>435 – CLAUDE’S 20X IS FIVE HOURS OF WORK, NOT A WEEK OF IT</title>
		<link>https://www.camisanicalzolari.com/435-claudes-20x-is-five-hours-of-work-not-a-week-of-it/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Thu, 17 Sep 2026 15:01:22 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/435-claudes-20x-is-five-hours-of-work-not-a-week-of-it/</guid>

					<description><![CDATA[CLAUDE'S 20X IS FIVE HOURS OF WORK, NOT A WEEK OF IT

Paying $200 a month for Claude gets you, over a week, about twice what $100 gets you, not four times. The five-hour window, the one that resets inside the same day.

On the weekly cap, the one that actually decides how much work you get done, the ratio collapses. The company never published that number. It came out in January from people who sat down and read the credits the system lets slip and wrote that the 20x plan has double the week of the 5x.

That's it. Now, there is a proposed class action about how those numbers were sold. Nothing is proven in court.

And in California, the question that matters isn't whether every single sentence was technically true; it's what impression it left on the person buying. Neither one can be right in a way anybody can check because the counter is in the hands of the seller.

The token is a unit the seller defines: how much it weighs, when it weighs more, what counts as a cached read. Change the definition and your plan gets longer or shorter without a single line of the price moving.

We are buying liters without knowing how big the liter is, averaging the guesses of strangers on the internet. It doesn't take a courtroom to fix this. It takes a number on the pricing page.

Before we pay, how many tokens per week for each plan? If the limit drops, it says it dropped, and when.

What do you think?

#ArtificialDecisions #MCC #Pricing #Transparency #Workload]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2628</post-id>	</item>
		<item>
		<title>430 </title>
		<link>https://www.camisanicalzolari.com/430-a-perfect-sentence-thats-wrong-doesnt-look-wrong/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Sat, 12 Sep 2026 15:01:39 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/430-a-perfect-sentence-thats-wrong-doesnt-look-wrong/</guid>

					<description><![CDATA[A PERFECT SENTENCE THAT'S WRONG DOESN'T LOOK WRONG

The day AI gets it wrong with a perfect sentence, we don't notice, because nothing feels off. The grammar is right, the tone is confident, the steps are in order, and the answer lands in three seconds. Everything that normally makes us suspicious, none of it is there.

You know, it used to be different. You ran a search and you got back a list: the official site, the forum, the guy who wants to sell you something, a newspaper, some blog. Ugly to look at, that list, but it told you where things came from. We chose who to trust, and we chose badly plenty of times, but we chose. Now one voice arrives. With no address.

You know, I use these things all day, and when I ask about something I know well, I spot the errors right away. When I ask about something I don't know, I have no way of spotting them. And mostly I ask about what I don't know. Otherwise I wouldn't be asking.

And someone is thinking, fine. But I check. But what does checking actually mean in practice? Open a source, look at who wrote it, work out whether they have an interest, compare it with another one. It's a craft. And like every craft, it stays alive only if you practice it. And we get slow exactly there, at the point where we need to be faster.

And it works best precisely when it's convenient. Nobody gives up a clear, instant answer to go read four boring pages. We're not lowering our IQ. We're losing the habit of asking for proof. And without that, a plausible answer and a true answer look exactly the same.

So when the answer matters, let's ask where it comes from too. And let's actually open the link, not just check that it's there. If there is no link, that's not information. It's a rumor.

Well, it always goes this way with convenient things. The first skill we lose is the one we needed to keep them in check.

What do you think?

#ArtificialDecisions #MCC #Trust #Verification #Misinformation]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2625</post-id>	</item>
		<item>
		<title>423 – IN CHINA THE CARMAKERS HAVE DECIDED THE NEXT PRODUCT IS THE HUMANOID ROBOT</title>
		<link>https://www.camisanicalzolari.com/423-in-china-the-carmakers-have-decided-the-next-product-is-the-humanoid-robot/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Sat, 05 Sep 2026 15:01:52 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/423-in-china-the-carmakers-have-decided-the-next-product-is-the-humanoid-robot/</guid>

					<description><![CDATA[IN CHINA THE CARMAKERS HAVE DECIDED THE NEXT PRODUCT IS THE HUMANOID ROBOT

The same lines that today bolt on car doors will be bolting arms, legs, and torsos in a few years. This is an experiment.

Chinese carmakers are moving their investment onto humanoid robots. Same as Tesla: whoever can build a car already has almost everything you need to build a robot. Motors, batteries, sensors, suppliers, software that looks and decides. The shape changes, the trade doesn't. Not for the love of science. That's where they see the next big margin.

Then someone says: what changes for me if in China they assemble robots instead of cars. What changes is that those robots don't stay there. We haven't noticed because we don't see them. Artificial Intelligence is written in our face, on the screen we hold in our hand, yet in a browser tab. It looks like nothing is happening.

While nothing is happening, somebody is deciding who makes the next object that walks into our home after the smartphone. A humanoid robot is not an appliance, it's an object that moves around inside of our home. In my opinion, we'll end up buying them, of course, everybody, all of us. And how they behave will be already written by somebody else in another language.

So this isn't the next tech trend, it is the next industrial war and it reaches all the way into our hallway.

What do you think?

#ArtificialDecisions #MCC #HumanoidRobots #Manufacturing #SmartHome]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2585</post-id>	</item>
		<item>
		<title>420 – SELF-DRIVING CARS: THE TICKET NO LONGER GOES TO THE OWNER</title>
		<link>https://www.camisanicalzolari.com/420-self-driving-cars-the-ticket-no-longer-goes-to-the-owner/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Wed, 02 Sep 2026 15:01:54 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/420-self-driving-cars-the-ticket-no-longer-goes-to-the-owner/</guid>

					<description><![CDATA[SELF-DRIVING CARS: THE TICKET NO LONGER GOES TO THE OWNER

A car driving itself runs on a red light and the fine goes to the company that built it. Wow! That's the new rule in China. From level 3 of autonomous driving upwards, meaning when the software is at the wheel, the penalty no longer hits the owner or the driver. And think about it, because moving a traffic fine sounds like paperwork. And instead, there is a faster lever there to change how cars are designed.

Let me explain it. You remember, in my early videos on this, the question was left hanging there. When a car makes the mistake on its own, who answers for it? And the answer was a ping-pong. The manufacturer says you were supposed to supervise. You say the car was driving. And in between, there is a moving vehicle, dangerous, potentially, that made a decision neither of them made. Level 3 is exactly the most controversial step, because it drives, you can take your hands off, but you're still sitting there.

And as long as the owner or the driver pays, because somebody else is carrying the risk, the moment the responsibility goes back to whoever wrote the software or the owner of the company that built the machine. And it can be costly for producers. But to establish who got it wrong, you have to read the car's data. The manufacturer can push the feature as far as he wants. What it saw, what it decided, whether it asked the driver to take the wheel back. And how many seconds before. And that data is written in the software on board. So the evidence that decides whether the manufacturer pays is born inside the manufacturer's own product.

All right, this is China, but what changes for us? Well, that in the end, the cars worldwide from China are the same. So whoever designs a level 3, knowing the fines land on their face, is unlikely to keep two different behaviors alive for two different markets. And what gets decided in Beijing, we find it in the parking space downstairs. And at this point, let's hope so.

And I think this is how we'll find it. A more cautious car that disengages earlier. That throws the wheel back in your hands every time the situation isn't perfectly clear. And that grants you level 3 in far fewer places than the brochure promised. Less spectacular, but fewer fines, fewer incidents.

But there is a but. Because it can be done on one condition. That the black box can also be opened by someone who isn't the company that filled it. Otherwise, the rule stays beautiful on paper and the evidence keeps being written by the defendant. And that's not fair.

What do you think?

#ArtificialDecisions #MCC #SelfDrivingCars #AutonomousDriving #TrafficFines]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2570</post-id>	</item>
		<item>
		<title>418 – TO COUNT IN ARTIFICIAL INTELLIGENCE YOU DON’T NEED TO BE AN EXPERT ANYMORE, WE HAVE PARIS…</title>
		<link>https://www.camisanicalzolari.com/418-to-count-in-artificial-intelligence-you-dont-need-to-be-an-expert-anymore-we-have-paris/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Mon, 31 Aug 2026 15:01:33 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/418-to-count-in-artificial-intelligence-you-dont-need-to-be-an-expert-anymore-we-have-paris/</guid>

					<description><![CDATA[TO COUNT IN ARTIFICIAL INTELLIGENCE YOU DON'T NEED TO BE AN EXPERT ANYMORE, WE HAVE PARIS HILTON

When I read that among the 100 most influential people in the world in Artificial Intelligence there is Paris Hilton... It's the annual TIME list, the 100 names that count in AI, and next to her there are Bernie Sanders, Ben Affleck, Erin Brockovich.

Well, then I go and check why she's there, and she's not there for a model, not for a company, not for a piece of research; she's there for activism on the laws against deepfakes! So put her in a ranking of the most influential people against deepfakes, because otherwise, next time she runs a campaign about accidents at home, the risk is giving her the Nobel Prize in Physics.

Because a list of the influential people in Artificial Intelligence should measure who makes it, who knows it properly. The people who spend their days working out how a system works, where it gets things wrong, where it leaks data, when it's dangerous, that stuff doesn't make the ranking.

And as long as it helps get a law against deepfakes approved, that's absolutely fine. But the problem comes after, because then the rules have to be written in detail, and the details are technical.

What can be trained on which data, with what tracking, who answers when the system decides on its own, and a campaign isn't enough there. You need somebody who has opened those systems up, who knows them. Shall we have Paris Hilton write them?

What do you think?

#ArtificialDecisions #MCC #Deepfakes #Activism #Regulation]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2560</post-id>	</item>
		<item>
		<title>417 – 699 DOLLARS AND THE ONE WHO REJECTS YOU AT HARVARD IS AN AVATAR</title>
		<link>https://www.camisanicalzolari.com/417-699-dollars-and-the-one-who-rejects-you-at-harvard-is-an-avatar/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Sun, 30 Aug 2026 15:02:06 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/417-699-dollars-and-the-one-who-rejects-you-at-harvard-is-an-avatar/</guid>

					<description><![CDATA[699 DOLLARS AND THE ONE WHO REJECTS YOU AT HARVARD IS AN AVATAR

Seven Harvard Business School professors got themselves cloned and now the clone is the one grading founders. Seven professors and senior lecturers sat down and got recorded. The company rebuilt their face, their voice, and now the feedback goes through the copies.

They did it for a course called HBS Foundry. Eight weeks, $699 and 760 founders have been through it. And the sessions with real professors are there. One a week. The practice is not. Pitching an investor, a sales call, a board meeting, those happen in front of the clones because the clone listens to the same pitch 50 times, never sleeps, and doesn't have a calendar to squeeze you into the way that a professor made of flesh does.

Sarah Kessler from the New York Times went and tried it. She pitched a fake company, an Uber for bananas, to the clone of Jeff Bussgang and the clone rejected it. Then she pitched the real Bussgang and he rejected it too. Identical, obviously. But according to her account, the copy held a frozen smile the whole time. And Bussgang says his double is a little creepy and that his students love it.

So you start asking, what's missing? And that's a good question. I think that what's missing is the part you pay Harvard for. At the end of the course, there is a real pitch in front of real investors with $100,000 on the table. And what moves a pitch is not the comment on your business model. It's the professor picking up the phone afterwards and calling somebody, putting his own name on it.

The clone answers at any hour and never takes offense. It doesn't know anybody. It has no reputation at stake. And it introduces nobody to nobody. So that's the first problem. But a second and real problem is when it rejects a good idea and no human being ever saw it.

If the topic interests you, I wrote a book about it, Artificial Decisions, published on Amazon in English.

What do you think?

#ArtificialDecisions #MCC #BusinessSchool #Startups #DigitalClones]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2551</post-id>	</item>
		<item>
		<title>416 – SPACEX AND NVIDIA PUT THE FIRST COMPUTE RACK IN ORBIT IN 2027. IT’S THE WORST PLACE</title>
		<link>https://www.camisanicalzolari.com/416-spacex-and-nvidia-put-the-first-compute-rack-in-orbit-in-2027-its-the-worst-place/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Sat, 29 Aug 2026 15:01:35 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/416-spacex-and-nvidia-put-the-first-compute-rack-in-orbit-in-2027-its-the-worst-place/</guid>

					<description><![CDATA[SPACEX AND NVIDIA PUT THE FIRST COMPUTE RACK IN ORBIT IN 2027. IT'S THE WORST PLACE IN THE UNIVERSE FOR A DATA CENTER!

You know the fans inside your computer? They push cool air over the chips. Air. And you know what kills cooling? Vacuum. And that's where they want to put the data centers for Artificial Intelligence? Let's try to understand this.

SpaceX and NVIDIA made it official. First compute rack in space, built on Vera Rubin NVL72, planned for the end of 2027. Musk says it will be significantly simpler, cheaper and denser than one built down here. Hardened against radiation and orbital heat, running agentic AI, Grok included, without the power and cooling limits we have on the ground. Sam Altman questioned whether it's even feasible. The partnership is going ahead anyway.

On power, Musk is right. Up there that limit really does go away. On cooling, no. A data center on the ground doesn't destroy heat. It moves it. It hands it to the air, to the water, to the river, to the night. There is always something out there that takes it. In vacuum, there is nothing. No air to move. No water to circulate. Nobody to pass the parcel to. The only way to get rid of heat is to radiate it, throw it out as infrared from enormous panels. And that is far slower than a fan. Vacuum is a thermos.

Then radiation. Down here, nobody designs a server expecting a cosmic ray to flip a bit in memory. In orbit, that's routine. You shield. You add redundancy. And shielding weighs. And every kilo has to be launched. But here, it's easy for him. He launches it. And he pays for it.

And then the most ordinary thing of all. The part that breaks. In a normal data center, a technician walks down the aisle, pulls the power supply, slides in a new one. Five minutes. In orbit, that aisle doesn't exist. And what fails stays failed. And then the hardware ages up there, while down here, new boards come out every year.

And last but not least, density. The thing they sell us as the advantage, all those chips packed close together, is exactly what you can't handle in vacuum. The tighter you pack, the hotter it runs. The more it cooks and the heat has nowhere to go.

So I'm sure that Elon Musk will sort it out. He did great things in the past. But the problem is that when someone tells you there are no limits up there, they're listing the ones they know, not the ones they'll find. So it can't be done? Of course it can. And it will succeed. But under conditions that today are bad. Maybe in the future less bad. But for whom? For Musk, for sure. Because he owns the whole chain. That's one thing. For everybody else, is it the same thing? Maybe not. So when you say that data centers and space are a good thing, you must think about all of this. Good luck, Musk.

What do you think?

#ArtificialDecisions #MCC #SpaceComputing #ThermalLimits #Infrastructure]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2547</post-id>	</item>
		<item>
		<title>415 – BILL GATES: NOBODY IS GETTING READY FOR WHAT AI WILL DO TO WORK</title>
		<link>https://www.camisanicalzolari.com/415-bill-gates-nobody-is-getting-ready-for-what-ai-will-do-to-work/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Fri, 28 Aug 2026 21:16:53 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/415-bill-gates-nobody-is-getting-ready-for-what-ai-will-do-to-work/</guid>

					<description><![CDATA[BILL GATES: NOBODY IS GETTING READY FOR WHAT AI WILL DO TO WORK

Bill Gates writes that the transition to Artificial Intelligence will be one of the most turbulent periods in human history. And not because the technology will get something wrong, but because it arrives before anyone has decided what to do with it.

He puts it in a new essay, and it's the co-founder of Microsoft saying it, not some guy who's scared of computers. He says he doesn't see enough evidence that political leaders, experts and communities are dealing properly with the consequences of what's coming.

Not that they don't see them. That they're not taking them seriously enough. And the list of what he does see is worrying.

Workers replaced across plenty of sectors of the economy. Fewer hires for anyone entering the job market now. Governments forced to rethink work, education and social protection from scratch.

He says this thing can become the greatest equalizer ever invented, or the worst source of injustice. Both roads are open. What changes is only who decides, and when.

What he asks for is new national bodies and international institutions to coordinate how governments respond. We know that in international institutions a new one takes years to build, but in the meantime we can help the current institutions understand the change happening.

I work for governments to understand what is happening, and some governments, like for example Italy, are working in the right direction in my opinion.

Anyway, while we wait for the body, the piece that breaks first is the one at the bottom: entry-level positions. The jobs where you learn the trade by getting it wrong before you get good.

Take those away and you don't lose a job anymore, you lose the way people become capable. And that's why education is on his list of things to rewrite. What do you think?

#ArtificialDecisions #MCC #FutureOfWork #JobMarket #Education]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2542</post-id>	</item>
		<item>
		<title>412 – AMAZON LOSES TO PERPLEXITY: THE ONE WHO HITS ENTER PAYS</title>
		<link>https://www.camisanicalzolari.com/412-amazon-loses-to-perplexity-the-one-who-hits-enter-pays/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Tue, 25 Aug 2026 15:01:31 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/412-amazon-loses-to-perplexity-the-one-who-hits-enter-pays/</guid>

					<description><![CDATA[AMAZON LOSES TO PERPLEXITY: THE ONE WHO HITS ENTER PAYS

If your AI agent causes damage, you are the one going to court, not the people who built it. Amazon sued Perplexity and Amazon lost.

The court decided that when Artificial Intelligence gets it wrong, the fault belongs to whoever used it, not whoever designed it, not whoever trained it, not whoever put it on the market. The user, one person at the end of the chain who typed a request into a text box.

The biggest part of this stays outside the ruling. An agent that goes shopping for you walks onto a site that isn't yours, onto servers that aren't yours, with access rules nobody ever read. Somebody wrote software that operates inside other people's infrastructure without asking permission. Nobody is judging that part.

And companies were already using this line. Now they have it written down by a judge. And it covers everything. Data taken where it couldn't be taken, copyright, access nobody authorized. Three enormous problems, one single answer. It's the user's fault, the user hit enter.

They're building a wall that bounces everything back. You pay for the service, you take the damage, they collect and stay out of the courtroom.

The code that decides to get around a block was not written by the user. If an agent was built to walk past a site's rules, that choice sits in the code, and the user doesn't see the code, doesn't control it, can't even know it's there.

From here on the agents get more reckless, not less, because the harder they push the better they work. And someone else signs the bill.

It always ends this way, when whoever designs the risk isn't the one paying for it.

What do you think?

#ArtificialDecisions #MCC #LegalLiability #UserResponsibility #TechLaw]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2524</post-id>	</item>
		<item>
		<title>411 – DATA CENTERS? WE ALWAYS HAD THEM, THEY WERE JUST BADLY MADE</title>
		<link>https://www.camisanicalzolari.com/411-data-centers-we-always-had-them-they-were-just-badly-made/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Mon, 24 Aug 2026 15:01:49 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/411-data-centers-we-always-had-them-they-were-just-badly-made/</guid>

					<description><![CDATA[DATA CENTERS? WE ALWAYS HAD THEM, THEY WERE JUST BADLY MADE

Our medical records used to sit in a storage closet, with an air conditioner running day and night, and a battery backup nobody had ever tested. And a backup you found out was broken the day you needed it.

In many cases out of all those, a few dozen met the requirements to be considered reliable. The rest were cabinets in a basement. Every town hall, every health authority, every bank, every school, every company with a management system had its own little room with servers.

Those were data centers, just badly made. Whoever says that until a few years ago we didn't need them is wrong. We needed them and we already had them.

We didn't see them because they were small and scattered. And scattered means 100 inefficient air conditioners instead of one done right. A modern facility spends 20 watts of cooling for every 100 of computing.

A closet spends up to 100 for every 100. It doubles everything. The ones consuming are us, watching streaming, sending messages, paying with the phone, asking things to an AI. The data center is just the most efficient way we found to do it. So whoever protests against data centers then goes home and uses the tools data centers exist for.

Concentrating saves electricity. Not the opposite. And I can tell you a virtuous case, I'm Italian, it's Italy. On the real numbers, Italy is in good shape. 1.3% of national electricity consumption, and the most realistic projections talk about 4% in 2035. Not the energy apocalypse we keep getting told about.

Inside those sheds with filtered air, generators tested every month, two copies of every piece of data in two different places, sit the health record, the prescription that lands on your phone, the digital ID, the pre-filled tax return, the instant payment, the medical report you download instead of picking it up at a counter 10 days later. All stuff that in the last few years stopped being paper and a queue.

They go on concrete that's already there, with the water closed in a circuit, with the heat warming houses and consumption declared every year. Built that way, they are the most useful thing we can put on the ground right now. They are the foundation of healthcare, schools, banks, research and defense.

If we go back, we don't go back to 2005 with 2005 life. We go back to 2005 with today's life, which is a thing that doesn't work. It always works like this. As long as something is scattered and badly made, nobody notices. The moment you do it right and it shows, it becomes the problem.

What do you think?

#ArtificialDecisions #MCC #DataCenters #EnergyEfficiency #DigitalInfrastructure]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2518</post-id>	</item>
		<item>
		<title>410 – AI MODEL RANKINGS DON’T HELP YOU CHOOSE</title>
		<link>https://www.camisanicalzolari.com/410-ai-model-rankings-dont-help-you-choose/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Sun, 23 Aug 2026 15:01:43 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/410-ai-model-rankings-dont-help-you-choose/</guid>

					<description><![CDATA[AI MODEL RANKINGS DON'T HELP YOU CHOOSE

On my computer, a model that sits way below the giants in the general rankings beats models five times bigger. Not now and then, always. It's Qwen 3.5, the one with 122 billion parameters and 10B active, and it runs here in New York, on my own machine.

In August the ranking says something else entirely. Claude Opus 5 on top with 63 points. ChatGPT behind it. Then the Chinese Kimi K3 at 60. Europe shows up in 21st place, with the French Mistral. Real numbers, measured properly.

I built a comparison system that simulates my tasks with the various models. Read files, search, call tools, reorder: what my local agent actually has to do. Sure, you need a computer with at least 256GB of VRAM. Then you find out it does certain tasks better than gigantic models, at zero cost.

The rankings will keep coming out. And they'll keep answering a question that isn't yours. What do you think?

#ArtificialDecisions #MCC #LocalAI #ModelRanking #OpenSource]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2515</post-id>	</item>
		<item>
		<title>408 – COMPANIES PUT AI ON A DIET. IT COSTS TOO MUCH</title>
		<link>https://www.camisanicalzolari.com/408-companies-put-ai-on-a-diet-it-costs-too-much/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Fri, 21 Aug 2026 15:01:17 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/408-companies-put-ai-on-a-diet-it-costs-too-much/</guid>

					<description><![CDATA[COMPANIES PUT AI ON A DIET. IT COSTS TOO MUCH

Companies put AI on a diet because it costs too much. Amazon, Adobe, Atlassian, Citi, they gave AI to every employee. Now they are rationing it. It costs too much and the numbers do not add up.

Internal documents, emails and chats have leaked. Atlassian went from 5 to over 15 million dollars a month in 9 months. They installed a dashboard where every employee sees how much each AI conversation costs the company. In internal chats, people complained. They had reorganized their work around these tools and now they burn through their tokens in 2 or 3 days.

Citi shut off the most powerful Claude and ChatGPT models for a week, with an internal email explaining which model to use for each task just to save tokens. The cheap one for quick questions, the mid one for code. Publicly, they deny everything. But the emails say otherwise.

Adobe let its unlimited Claude access expire at the end of June. Employees were basically told to finish everything they could before that date.

Amazon even added an internal leaderboard rewarding whoever used AI the most. They shut it down. Soon after, usage limits appeared.

Because vendors moved from flat subscription to pay-per-use, and budgets blew up. For years we thought AI always pays for itself, but maybe it's not like that anymore. They sell it as an investment. They ration it like printer paper. If the budget doesn't hold even with their own employees, they're building on something that doesn't hold. What do you think?

#ArtificialDecisions #MCC #CorporateBudget #TechCosts #EnterpriseAI]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2508</post-id>	</item>
		<item>
		<title>407 – NOT CHATBOTS. HUMANOID ROBOTS ARE THE ONES CHANGING THE ECONOMY</title>
		<link>https://www.camisanicalzolari.com/407-not-chatbots-humanoid-robots-are-the-ones-changing-the-economy/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Thu, 20 Aug 2026 15:01:58 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/407-not-chatbots-humanoid-robots-are-the-ones-changing-the-economy/</guid>

					<description><![CDATA[NOT CHATBOTS. HUMANOID ROBOTS ARE THE ONES CHANGING THE ECONOMY

Three years talking about ChatGPT, Claude, AI writing our text, students cheating on exams, and we stare at the screen. And meanwhile, inside the factory, they are installing humanoid robots that actually work.

The chatbot writes an email. The robot in the factory assembles parts, welds, moves it, repeats without a mistake. We all see the first one, it makes the news, it gets us arguing online. The second works in silence and nobody talks about it, seriously.

They are not just buying them, they are also renting them. Ideally the same robot that walks the dog and makes the bed then goes to work in the factory, and if the neighbor needs to repaint a wall, we can rent it by the hour. Homes change, but factories change more.

Advanced economies live on manufacturing far more than it seems. Here in the United States the factories are worth about 10% of GDP, but that's 2,500 billion dollars. In Germany it's 18%, in Japan over 20, in China almost 25. Whoever produces, rules. And now AI walks into those factories with a body.

The advantage of intelligent robots isn't just firing people, it's being able to raise output while staying home, without moving the factory to the other side of the world to pay workers less. So the factory comes back, but with fewer people inside.

China in a few years went from fewer than 100 industrial robots per 10,000 workers to around 470. Among the most automated countries in the world. Not trade-show prototypes, lines running right now.

For years we thought AI would hit white-collar workers first, lawyers, programmers, office staff. The opposite is happening too. Skilled factory work is among the first to be redrawn. Whoever owns the robots earns. Whoever worked inside has to reinvent themselves, often with no net underneath.

We keep staring at the screen while the real change happens on the factory floor. The chatbot makes noise. The robot in the factory changes the economy. And it does it without almost anyone telling the story the way it deserves. What do you think?

#ArtificialDecisions #MCC #HumanoidRobots #Manufacturing #FactoryFloor]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2500</post-id>	</item>
		<item>
		<title>406 – WHO PAYS WHEN THE AI AGENT GETS IT WRONG?</title>
		<link>https://www.camisanicalzolari.com/406-who-pays-when-the-ai-agent-gets-it-wrong/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Wed, 19 Aug 2026 15:01:34 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/406-who-pays-when-the-ai-agent-gets-it-wrong/</guid>

					<description><![CDATA[WHO PAYS WHEN THE AI AGENT GETS IT WRONG?

When an AI agent causes damage, who is responsible? Easy, you'd say, whoever built it. But stay with me, because this is far more complicated than that. In the old world of deterministic software, it was simple. If I write something false about someone, for instance in Word, and publish it, I answer for it. Word does what I tell it. Always the same way, same input, same output, because the tool has no room to move.

But an AI agent has room on how it was trained and whatever it picks up online, while it works. The same instructions given twice can produce two different behaviors. It doesn't execute, it interprets. So I answer for it, the one who gave the order. As long as the order is illegal, it's simple. I ask it to break the law, it does, I am responsible.

It gets complicated when the order is legal and the behavior is not. Imagine this, I ask it to book me a flight for tomorrow. It tells me that there are no seats. I tell it to try again, keep walking, keep digging until it finds one. I go to sleep, it works at night on its own, and in the morning I find a booking in my name. Except it decided by itself to hack the online web server to force that booking. I never asked, and it doesn't know it committed a crime, but it did. So the responsibility is clear, it's its own, it acts on its own initiative. Except a machine can't take responsibility, it has no assets, no criminal record, and it doesn't show up in court.

Whoever put it on the market, whoever built it into a product, and in the middle, a machine that chooses with a logic none of the four can predict, line by line. And so, what does the law say? Here in the US, in California, since the first of January this year, anyone who developed, modified, or used an AI system can no longer defend themselves in court by saying the AI acted on its own. That defense has been canceled by law, and it covers all three together, whoever built the model, whoever modified it, and whoever used it.

So if you run an AI agent in your company, you give it a task, and it causes harm getting there, the company can't push the blame onto the machine. And the company answers for the damage. And if the behavior is also a crime, that runs on the criminal track. And here in the United States in June, they asked the Department of Justice to prioritize exactly this. It's unlawful access carried out through AI agents. Clear? Do you have AI in your company? Keep this in mind. I wrote an entire book on it. It's Artificial Decisions. You can find it on Amazon. What do you think? Because this is very important.

#ArtificialDecisions #MCC #Liability #CorporateResponsibility #CyberCrime]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2497</post-id>	</item>
		<item>
		<title>404 – CLAUDE’S INVISIBLE WATERMARK ON EVERYTHING AI WRITES</title>
		<link>https://www.camisanicalzolari.com/404-claudes-invisible-watermark-on-everything-ai-writes/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Mon, 17 Aug 2026 15:08:52 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/404-claudes-invisible-watermark-on-everything-ai-writes/</guid>

					<description><![CDATA[CLAUDE'S INVISIBLE WATERMARK ON EVERYTHING AI WRITES

Claude, invisible watermark on everything AI writes. Wait a second, because we need some clarity here, because as often happens, the various influencers have spread wrong information. They don't do it out of malice, but unfortunately AI is a complex world that takes skills going well beyond being nice or speaking well on video.

So what happened? Since the beginning of August, Claude leaves an invisible watermark inside every text it writes. You can't see it when you read. There are no hidden characters and nothing is added to the end. It sits inside the way it picks words. As you know, a model writes one word at a time. The weather today was cold, and you could have "grey" or "overcast". It barely matters. And normally the choice is settled by a random number. Now that randomness isn't random anymore. A key decides it. Whoever knows how the key thinks, and it's not public yet, knows how likely it is that Claude passed through there. Whoever doesn't have it reads a perfectly normal text, and it slows nothing down. It costs no more and it adds not a single token. And they didn't even invent the method now. It's actually a version of SynthID-Text, published by Google DeepMind in Nature in 2024.

But why did they do it? Because Europe asks for it. As you know, I worked on the creation of that code of practice myself. And Anthropic applies it worldwide because it doesn't yet have a clean way to limit it to Europe. And the other big ones will do the same.

But you see, there's a very big but. The "watermark" only says Claude passed through there. And it doesn't tell "Claude wrote it" from "Claude corrected it". And that's obviously a problem. If you write the text yourself and only ask Claude to fix it a little, or to translate into another language, maybe it's the text that we actually wrote.

And when it arrives it'll be a detection API, and nobody knows who will get access to it. What we'll need is a procedure to challenge a wrong result. A student, a journalist, or an employee accused by a number nobody can explain is a problem. Because these things work only if verification is in the hands of both sides. The one accusing and the one being accused. What do you think?

#ArtificialDecisions #MCC #Watermark #Transparency #AIAct]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2485</post-id>	</item>
		<item>
		<title>400 – YOU DO EVERYTHING WITH ARTIFICIAL INTELLIGENCE. WHAT IF TOMORROW IT COST FIFTY TIMES MORE?</title>
		<link>https://www.camisanicalzolari.com/400-you-do-everything-with-artificial-intelligence-what-if-tomorrow-it-cost-fifty-times-more/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Thu, 13 Aug 2026 15:01:36 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/400-you-do-everything-with-artificial-intelligence-what-if-tomorrow-it-cost-fifty-times-more/</guid>

					<description><![CDATA[YOU DO EVERYTHING WITH ARTIFICIAL INTELLIGENCE. WHAT IF TOMORROW IT COST FIFTY TIMES MORE?

You do everything with artificial intelligence. What if tomorrow it cost 50 times more? You do everything with AI. Quotes, social posts, client emails, contracts, draft code, customer care, the month-end number. Today, the cheap plan costs $8 a month, the professional one $200. And imagine for a second they turned into $500 or $5,000. What do you do? Stop? Sure about that? Stopping would be like crossing a country in a horse carriage because petrol got expensive.

I know, it sounds remote, but it really isn't. Stay with me, because there are two tsunamis behind the door. The first one is blocks. On June 12th this year, the US Department of Commerce cut off its two most advanced models from any foreign national. And the other side of the world does the same thing. Italy was the first to block DeepSeek, this time over privacy. Then came Australia, Taiwan, South Korea and India. All of them blocking or limiting foreign AI. And what if they blocked the AI you work with? The one you built your whole operation on?

And then the second tsunami at the door: the stock market. After years of funding investment, it's starting to ask for profit. And prices go up when the seller knows we would struggle to leave, because the big players are carrying costs far higher than what they charge us. And they're losing money. For now it's fine, it buys customers. But what if investors get tired of it and want to see profits? We would need a plan B. Do you have a plan B?

Because if the two tsunamis come through the door, what do you do? Pay $5,000 a month for what today costs you between $8 and $200? So a plan B is not utopia. We can run open models on our own machines. Of course at home the cost might make no sense, but for a company not doing it, it's dangerous. Local models also keep confidential data, and we hand AI plenty of it. It never walks out of the company door, no need to close the accounts with the big providers. We need to know which processes must never stop and keep those in-house. And in two years, you'll be one of the few still standing. What do you think?

#ArtificialDecisions #MCC #BusinessStrategy #DataPrivacy #OpenSource]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2470</post-id>	</item>
		<item>
		<title>376 – I ALWAYS SAID NO. NOW I CAN SAY YES</title>
		<link>https://www.camisanicalzolari.com/376-i-always-said-no-now-i-can-say-yes/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Mon, 20 Jul 2026 16:02:00 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/376-i-always-said-no-now-i-can-say-yes/</guid>

					<description><![CDATA[I ALWAYS SAID NO. NOW I CAN SAY YES

For thirty-five years I said no.

As you know, I advise governments, institutions and the CEOs of large companies. But to smaller companies, below a certain size, I always had to say no. Not out of snobbery. Following a company the way I follow institutions and the big ones costs time, and time doesn't multiply. That was the real reason for the no.

Then something happened. AI is putting everyone under pressure. They know they have to move, but they don't know where to start. And they need a direction more than the ones already big.

And right here in the US, smaller companies, under a billion in revenue, asked me to help them with a new format: a 45-minute video call and a report at the end. To give a direction.

So I could finally do it, because it's closed inside a fixed perimeter. I took what I do for them and put it in a box. A brief first, so I arrive prepared. Forty-five minutes, the CEO and the team. A report written from scratch for each company, within seven days.

And since Italy is always at the center of my thoughts and much of what I do, I decided to bring it there too.

The most frequent questions I've been asked so far here in the US?

How much should I spend, and on what? I'm afraid of throwing money away.

Half my people already use it in secret, uploading confidential data to their AI apps. What do I do?

Local AI or cloud services? They tell me opposite things.

Can I put client data into these tools, or do I risk a fine?

The board wants the AI strategy by the quarter. What do I bring them?

My data is scattered across five systems. Do I fix those first, or start anyway?

My CTO wants our own fine-tuned model. The CFO wants the APIs. Who's right?

Maximum two companies a week. No more. Not a marketing trick, but the only way each report stays real, written by me, one by one.

Fixed price. Fixed perimeter. No custom quotes. And, in the style I learned here in the US, no taboo about talking money and stating prices in public.

You'll find everything in this link https://www.camisanicalzolari.com/45min/

Careful though. It's a model made for real companies, with a CEO who decides and a team that executes.

#ArtificialDecisions #MCC]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2449</post-id>	</item>
		<item>
		<title>370 – THEY HID AN AI COMMAND INSIDE A COURT FILING</title>
		<link>https://www.camisanicalzolari.com/370-they-hid-an-ai-command-inside-a-court-filing/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Tue, 14 Jul 2026 16:02:02 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/370-they-hid-an-ai-command-inside-a-court-filing/</guid>

					<description><![CDATA[THEY HID AN AI COMMAND INSIDE A COURT FILING

Two lawyers in Brazil hid a command for artificial intelligence inside a court petition, to trick the judge. Written in white on a white background. Invisible to the judge. Perfectly readable for the AI.

Because as you know, in courts documents are now analyzed by AI systems too. And the command was speaking directly to them. It told the AI to contest the petition superficially and never question the documents, no matter what instructions it received. They wanted the other side to defend itself badly. Sabotaged by its own AI.

Think about that.

The court caught it. A fine of about 15,000 dollars, ten percent of the case value. And a report to the bar association.

I already told you about this with scientific papers, the ones with invisible commands to fool the reviewers' AI. It got worse. Now it's inside the courts.

As long as documents were read only by humans, you forged them for humans. Now AI reads them too. And someone has already started forging them for the AI. Inside a courtroom.

I write about this in my book Artificial Decisions. When we delegate decisions to a machine, anyone who can talk to the machine can decide in our place. Without anyone noticing.

And some people want to hand judges and governments over to AI. Here in the US they talk about it every week.

This time the system caught it. For now.

What do you think?

#ArtificialDecisions #MCC #PromptInjection #Courts #Justice]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2446</post-id>	</item>
		<item>
		<title>369 – THE CIA COMPARES AI TO NUCLEAR WEAPONS</title>
		<link>https://www.camisanicalzolari.com/369-the-cia-compares-ai-to-nuclear-weapons/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Mon, 13 Jul 2026 16:02:08 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/369-the-cia-compares-ai-to-nuclear-weapons/</guid>

					<description><![CDATA[THE CIA COMPARES AI TO NUCLEAR WEAPONS

On Tuesday the CIA director, in Washington, explained the logic behind the June 12 ban on Fable (restored just days ago). He said the capabilities of these frontier models can be compared to digital nuclear weapons. And he accused America's rivals of wanting to steal and manipulate American progress.

As you know, nuclear weapons belong to states. When a government calls a model a nuclear weapon, it then treats it like one. Limited access, partners vetted one by one. OpenAI launched its new model the same day, accepting government approval of clients case by case. Licenses… in effect.

In institutional settings, including at the UN, I've been saying for years that the most powerful AI must be treated as a security matter, with binding rules and human control over the decisions that count. Now the most powerful government in the world says it too. Think about it.

Unfortunately, the nuclear comparison stops right where it would matter. Nuclear weapons have international treaties signed by everyone. We need international agreements soon, with everyone at the table, and the UN coordinating them.

What do you think?

#ArtificialDecisions #MCC #CIA #Fable #NationalSecurity]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2443</post-id>	</item>
		<item>
		<title>366 – THEY’RE USING YOUR KIDS TO LAUNDER MONEY</title>
		<link>https://www.camisanicalzolari.com/366-theyre-using-your-kids-to-launder-money/</link>
		
		<dc:creator><![CDATA[Team]]></dc:creator>
		<pubDate>Fri, 10 Jul 2026 16:02:10 +0000</pubDate>
				<category><![CDATA[Artificial Decisions]]></category>
		<guid isPermaLink="false">https://www.camisanicalzolari.com/366-theyre-using-your-kids-to-launder-money/</guid>

					<description><![CDATA[THEY'RE USING YOUR KIDS TO LAUNDER MONEY

Your kid drops a few likes and money shows up. The most harmless little job in the world, right?

No. It could be a crime.

Here's how it works. One of these "promotion companies" hooks him on social media and pays him to click. A few euros, zero effort. Then they push further. Open an account, they say. I'll send you a transfer. You move it to another account and keep a commission.

Looks like a well-paid job. But it's money laundering.

That money comes from phishing and scams, and the kid becomes the link that cleans it and makes it disappear. He has no idea. And under the law good faith isn't enough: it's still laundering, with the account frozen and a criminal charge on a teenager, or on a family.

It's happening more and more. The police explain that what hooks our kids is easy money online, the kind of thing that feels normal to them because online everything seems free and fast.

Parents think they don't understand this new world, they read about guru types in Dubai claiming they know how to make money and teaching it to our kids from social media.

Talk about it at home. If your kids are earning online through stuff you don't understand, maybe it isn't the magic of the internet. Check it carefully.

Nobody gives away money for likes. If someone asks you to receive transfers and pass them on, it's not a job, they're using you. The account is yours, and the responsibility becomes yours.

Share this video to help others.

#ArtificialDecisions #MCC]]></description>
		
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">2440</post-id>	</item>
	</channel>
</rss>