Slovenian domain demand surges amid Trump’s AI rebrand

Welcome to The Signal from Superpower Daily with Maya and Theo Right and this is your weekly digest of the past week's most important AI developments We have a lot to cover We really do But first imagine placing a massive financial bet You are not betting on a company Right You are not betting on a stock You're betting that the entire global tech industry will blindly obey a single terminology memo from Donald Trump Which sounds crazy It does but that is exactly what is happening in Slovenia right now Let us jump straight into our lead story So it is a perfect collision of politics and digital

real estate I mean the demand for these web addresses is completely unprecedented Yeah we will look at the underlying numbers and we will also examine the psychological catalyst driving this whole market behavior So Slovenia manages the SI web domain and registrations for those addresses just exploded in September 2026 Exploded is the right word The organization managing these domains is called Register SI Their spokeswoman is Clara Herman She provided some staggering data points Right She reported 44 000 registrations in September alone Wow Yeah that is a massive jump You really have to compare it to the previous month August saw fewer than 2 000 registrations That

is just wild I know The pace also accelerated dramatically right at the end of the month Exactly 11 000 domains were registered on September 30 alone We do need to clarify something here though There is a slightly different count floating around the tech press Oh right the TechRadar one Exactly TechRadar cited a different tally They reported 46 066 September registrations and they compared that to 3 515 in August Which is still a huge jump It is But the TechRadar headline claimed an increase of over 100 times That headline is actually a bit misleading How so Well it compared one unusually busy day in September to

an average quiet day in August Ah I see Right It did not compare the total monthly volume But I mean either way you slice the data the numbers show a sharp and undeniable jump The trend line just went vertical Wait let me stop you there because we need to explain the catalyst behind this vertical line Yes the why Right This did not happen in a vacuum Donald Trump issued an executive order The order mandates a strict change in terminology for United States federal agencies This is the crazy part Yeah These agencies must entirely replace the term artificial intelligence in their official communications The new mandated

term is super intelligence Super intelligence Right And the acronym for that is SI Exactly So buyers are rushing to buy the SI domains They basically hope this SI acronym will become the new global standard What is fascinating here is the market psychology I mean this is speculative domain investing in its purest form Totally It is very important to differentiate this behavior from malicious cyber squatting Right Because that is different Very different Cyber squatting involves registering specific trademarked names in bad faith You know the goal there is usually extortion Like holding a brand hostage Exactly You register a brand name and hold it hostage You might

also use it for impersonation or phishing But this Slovenian surge is entirely different They're just grabbing everything Right These buyers are taking a macro level gamble on a new label They are reserving thousands of generic names They are banking on the hope that aggregate demand will just rise overall Because they want to flip them Exactly They want to sell them later on the secondary market for a massive profit You know this domain rush feels exactly like the late 90s dot com bubble Oh absolutely You are basically buying a piece of digital paper Like buying the naming rights to a star Does it actually mean anything

Right I seriously question if this is a rational investment It feels like a pure unadulterated gamble I mean the entire bet is based on a political memo Yeah And that memo holds absolutely zero weight in Silicon Valley Why would a private startup care about an internal government style guide That is the crucial limitation of this entire phenomenon You also have to look at the underlying economics The Slovenian registry charges registrars exactly 10 euros per domain Okay So there were wild claims circulating on social media People said this rush will bring millions of euros to Slovenia Millions Yeah but the registry entirely denies those claims They

explicitly stated the claims have no real basis Right If you do the math 44 000 domains at 10 euros is 440 000 euros Which is nice It is a nice revenue bump Sure It is certainly not a macroeconomic windfall for the entire country Right But the bigger caveat is the actual scope of the executive order itself Trump's directive only dictates terminology for U S executive branch agencies Just the executive branch Exactly It applies to their official correspondence It applies to their internal reports and procurement documents It absolutely does not apply to private businesses No It does not apply to the wider global tech industry The

government is essentially just changing its internal dictionary Right That does not legally force any company to rebrand its products That is true legally But we have to consider the power of government procurement Okay that is a fair point Right If a major defense contractor wants to sell software to the Pentagon they might adopt the language of the buyer Oh just to get the contract Exactly They might start calling their products superintelligence just to align with the federal procurement requests That is the sliver of logic driving these domain investors I still struggle to see Google or OpenAI abandoning a 70 year old term though Yeah it

is tough to imagine I mean the term artificial intelligence has incredible brand equity It dates back to the Dartmouth conference in the 1950s You do not just erase that overnight Exactly Which raises the most important question for those investors Will private sector adoption of the SI acronym actually materialize Right That is what you must watch next The entire gamble relies on the broader private sector organically adopting the government's terminology Which is a huge if Huge If the wider tech industry stubbornly sticks with the term AI well these domain bets will be completely worthless Just dead money Right But if the industry slowly shifts to SI

those addresses could become highly valuable digital real estate It is a high stakes waiting game We will explicitly close the book on the Slovenian domain rush right there So in other news we are transitioning from the hype of government level AI rebranding to a very grounded corporate reality Yeah We are looking at Mercor They just published a study where frontier AI models scored perfectly on simplified month end accounting scenarios This is our first top story Right The Mercor research dropped on October 1st Right And the AI models completed the accounting work significantly faster than human accountants They were also perfectly accurate Completely perfect Yeah The

models achieved a 100 score The human baseline in this study is incredibly important to point out Because they were not just interns Exactly The researchers did not use entry level bookkeepers They used 12 fully licensed certified public accountants Actual CPAs Yes The CPAs averaged five and a half years of professional experience And the AI models beat them comprehensively Wow The models were also vastly more efficient economically The cost per grading criterion met was more than 10 times cheaper for the AI compared to the human accountants That is a massive difference It really is This demonstrates the immense raw capability of AI in specific environments We

are talking about highly structured file search tasks here Right The assignments required finding exact figures within provided documents The models then had to calculate results based on those figures Finally they had to deliver a perfectly formatted table It is very rigid Yeah But the rapid progression of this technology is staggering Just 18 months prior the very best models completely failed this exact type of test Completely failed They scored below the human average of 37 Now just a year and a half later they are hitting perfect 100 scores Here's where the mechanics of the study get really interesting though The perfect AI scores actually ruin the

original scientific goal of the research Oh really Yeah The researchers initially set out to measure human AI collaboration They wanted to see exactly how much an AI assistant improves an experienced accountant's workflow Okay But the models hit the absolute score ceiling entirely on their own There was zero room left to measure any human improvement Because you cannot score higher than 100 Exactly The tool essentially became the entire worker You know think of this like passing the written part of your driving test Oh that is a good analogy Right The AI knows all the rules perfectly on paper It can find the figures and calculate the

tables in a perfectly quiet room Right But that is very different from actually driving a car in chaotic unpredictable city traffic I really have to push back hard on the framing of this study A perfect score on a highly sanitized test is impressive but it absolutely does not equal the ability to run an actual corporate finance department Real accounting is incredibly messy That is the fundamental limitation of this Merkur study You really have to look closely at the stripped down working conditions It was too perfect Exactly The test intentionally left out massive complicated parts of the actual job The simulated environment was entirely sterile No

office drama Right There were no confused client conversations The accountants had no co workers to message for clarification Which is huge Yeah They lack the deep accumulated company context that a human builds up over years of employment The tasks were completely isolated and perfectly packaged So if you look at the broader APEX DASH accounting benchmark instead it paints a very different picture A much more realistic one Exactly The full APEX benchmark is much closer to reality It contains 160 complex tasks spread across 10 simulated companies Right The work includes reconciling messy accounts and accruing ambiguous expenses And Claude Opus 5 5 was the leading model

on that benchmark And how did it do It only met 61 8 of the grading criteria That is a massive drop from a perfect score Yeah If we connect this to the everyday reality of corporate finance the failure rate is incredibly telling For sure Almost 60 of the tasks on the full APEX benchmark were fully unsolved by any model Right Meeting some grading criteria is absolutely not the same as finishing the job The AI can handle beautifully structured self contained spreadsheet work But it completely struggles with complex multi step reconciliations that require hunting down discrepancies across a whole company's fragmented software ecosystem So you should

watch whether future model updates can actually master those wider messy responsibilities Exactly The simplified assignments intentionally left out the chaotic context of a real business Watch closely to see if AI can start handling ambiguous unstructured financial data without human hand holding That is the real threshold for automation So next up while AI is desperately trying to master these complex human corporate tasks the AI companies themselves are facing a harsh truth Which is what They are realizing they desperately need highly trained humans to actually deploy these models into legacy businesses Yeah We are looking at Anthropic for this top story They are committing 100 million to

train 10 000 engineers for cloud deployments That is a massive budget It is They are officially launching the Cloud Frontier Academy The stated goal is to hit that 10 000 engineer mark by the end of 2027 And the first physical cohorts are already operating right now They are meeting in San Francisco New York and London The initial groups actually feature engineers from major established consulting firms Oh really We are seeing participation from Accenture and Deloitte Morgan Stanley is also deeply involved That makes sense This massive investment reveals a critical deployment bottleneck in the entire AI industry I mean building a hyper intelligent frontier model is

only half the battle You then have to integrate that model into a legacy corporate system That system might be running 20 year old databases Which sounds like a nightmare It is That integration requires highly specific deeply technical human skills So the goal of this academy isn't just to teach basic API coding Anthropic wants to create a whole new job category They are calling them Frontier Deployed Engineers These specialized engineers will shepherd a cloud project from the very first meeting to the final launch End to end Right They will handle the initial use case selection They will manage the grueling corporate security review They will drive

the final technical implementation It is a lot of work It represents a massive shift in focus The industry is really moving from simply building AI To the grueling work of integrating AI I like to compare the structure to a medical residency Oh how so Well doctors learn by actually doing the work right They work through complex patient cases under strict supervision Anthropic is explicitly leaning into this exact educational model The program participants first practice a simulated enterprise deployment in a safe environment Then they must pass a rigorous practical assessment Only after passing do they enter a mandatory 12 week real world residency If you have

ever tried to integrate a new software tool into your company you know it is a complete nightmare Oh absolutely So I am warmly skeptical of Anthropic's true motives here On the surface this looks like a genuine educational initiative Right It aims to help the broader industry overcome a skills shortage But is it actually just a highly funded brilliant strategy to create a locked in army of enterprise consultants That is the big question I mean these newly minted integration experts will naturally default to using CLAWD for every single client they advise Because that is what they know Exactly They will be exclusively trained on Anthropic's specific

software ecosystem It looks like the ultimate enterprise sales pipeline That is a very fair and realistic critique We also need to look at the strict limitations of the program itself This 10 000 figure is a future goal It is an ambitious target for 2027 It is absolutely not a count of qualified people working in the field today Good point The entry requirements for the academy are also incredibly strict This program is strictly nomination only Oh So you cannot just sign up Nope It targets highly experienced software engineers It is not an open introductory coding boot camp for beginners looking to switch careers And the requirements

to actually get the final credential are intense You do not just show up to lectures Participants must successfully lead a named real world CLAWD project This project must be executed at their own organization Wow They do get technical support from Anthropic during the integration process But then they must pass a final comprehensive assessment So perfect attendance alone does not earn the deployment badge Exactly You have to prove you can ship a working product Listeners should closely watch early 2027 That is exactly when the first wave of final credentials will actually be awarded to these engineers We will finally see if this rigorous medical residency model

for AI deployment actually works at a global scale The enterprise software industry needs these integration experts badly They really do Because the models are useless if they cannot talk to a company's existing data Absolutely Meanwhile Anthropic is spending millions of dollars teaching humans how to work with AI But other companies are intensely focused on teaching AI how to seamlessly pass as human Yes Next up we are analyzing Tavis They just previewed a new video call AI that passed for human in 48 percent of a small company test Forty eight percent Almost half The new system is called Griffin Light The preview took place on October

1st And the results from their internal testing were genuinely striking Exactly 26 out of 54 participants mistook the AI avatar for a real living human being And that was during a one minute interactive video call That represents a massive sudden leap in performance I mean we can compare this directly to their previous generation system Right The old system fooled only one in 41 people That was a success rate of just 2 4 percent So from 2 percent to almost 50 Exactly Jumping from 2 percent to nearly 50 percent in one generation is incredible Griffin is not just a static deep fake face mapped onto a

video either Tavis specifically calls it a human interaction model The underlying technology is fascinating It processes multiple complex streams of data in real time Like what Well it analyzes the user's speech and tone of voice It constantly watches facial expressions and hand gestures Wow It even tracks conversational pauses and breathing patterns Then it generates an overlapping continuous physical exchange Overlapping Right The AI avatar might subtly nod its head while you are speaking It visually reacts to what is happening in the conversation before it even formulates a verbal reply It mimics active listening You know I think of this phenomenon as the first date illusion The

first date illusion Yeah Almost anyone can seem perfectly normal charming and highly attentive for 60 seconds That is very true I seriously challenge the idea that fooling someone for one single minute means the technology is fully ready for the real world Right Does this brief illusion translate to a stressful 20 minute customer service call about a lost package Probably not Exactly Can this avatar maintain a coherent consistently helpful persona during a complex hour long tutoring session You are hitting directly on the primary limitation of the study You have to consider expectation bias here Oh The test subjects explicitly expected to be meeting a human being

on the call That strong expectation heavily influenced the final results They were not looking for an AI Exactly They were not actively looking for digital glitches or unnatural eye movements We also have objective data from the NVIDIA Video FDB benchmark to consider Okay GryphonLight scored 3 83 out of 5 for generating realistic conversational behavior Which is good It is That is actually very close to the 3 92 human reference score It generates behavior beautifully But the perception score tells a completely different story Yes it does Generating a nod is easy The AI actually has to understand the conversational cues to know exactly when to nod

Right GryphonLight scored a much lower 3 73 in perception The human reference score for perception was 4 20 That is a big gap It is It looks convincingly human on the surface but it does not accurately read human emotions as well as a real person does Yeah It is also highly important to note that GryphonLight remains strictly restricted to trusted testers There is currently no public launch date That brings us to exactly what listeners must watch next TAVIS must handle the safety and disclosure protocols incredibly carefully Oh absolutely How will they legally and ethically inform users they are speaking to an AI They also must

absolutely prevent unauthorized identity copying before a wider public release Because of defakes Exactly If anyone can clone a face and voice the potential for catastrophic misuse and fraud is significant As these AI systems become incredibly powerful and visually lifelike securing the physical hardware to run them has escalated into a high stakes global game It really has It is sparking international black markets and smuggling rings So moving on we are examining a major federal law enforcement action The U S Justice Department arrested the CEO of EarthMade Computer Right The arrest is over alleged smuggling of more than 300 million in export controlled NVIDIA servers 300 million

Yeah Wow Yeah The CEO's name is Greg Liu The federal allegations are severe Prosecutors claim he orchestrated a massive highly sophisticated smuggling scheme And this went on for a while Yes The illicit operation allegedly ran continuously from October 2023 all the way to August 2026 The sheer financial scale of this alleged operation is staggering The government claims Liu successfully smuggled over 300 million worth of highly restricted tech into China That is a huge amount of hardware It is And the logistics were incredibly complex The shipments did not go directly from California to Beijing No they never do Right They passed through multiple layers of intermediaries

They routed the physical hardware through Singapore They used obscure freight forwarders in Malaysia Wow They moved the goods through shell companies in Hong Kong The final destinations were tech hubs in places like Hangzhou We actually have a very specific documented example directly from the federal indictment Oh what happened There was a major 2024 order for 27 complete server racks The total value was roughly 7 614 million The official export paperwork falsely listed Malaysia as the final destination However a cooperating co conspirator later admitted the servers were ultimately sent directly to China So they just lied on the forms Basically This specific case highlights the severe

ongoing enforcement risks surrounding this hardware right now You know U S export controls feel exactly like a hopeless game of global whack a mole They really do A motivated buyer creates fake shipping documents They invent a fake corporate entity with a fake CEO named Jackie Louie Right Jackie Louie They use a labyrinth of freight forwarders in Malaysia who ask zero questions How can any U S tech company realistically police this supply chain It is almost impossible Exactly How can a federal government agency physically track where a server rack ultimately ends up once it leaves a U S port and enters the global shipping network That

is the core Seemingly unsolvable challenge of global tech embargoes The hardware is highly restricted for a very specific reason Right These massive servers contain thousands of A100 and H100 processing systems These specialized graphics processors are absolutely essential for training the next generation of large language models The real frontier stuff Exactly The United States government has deep existential fears about Chinese A I advancement They are specifically worried about the models being used for advanced military applications and cyber warfare The logistical loopholes to bypass these controls are vast and highly complex We must highlight a crucial legal caveat here though These are purely unproven allegations at this

early stage Yes Very important to note We has been formally arrested but he absolutely has not been convicted of any crime Right The potential penalties are severe He faces up to 20 years in federal prison for conspiracy He faces another potential 20 years for money laundering Well the specific smuggling count carries up to 10 years The Justice Department is also aggressively seeking to seize over 176 million dollars in alleged illicit proceeds from the year 2024 alone NVIDIA also officially released a public statement regarding the situation They stated that less than zero point five percent of their total products are diverted to China overall OK however

that is a very broad aggregate figure It addresses their estimated global diversion rates across all product lines Right It absolutely does not address the granular specifics of Louis's alleged multimillion dollar shipments Listeners should watch the federal courts closely on this one Watch whether federal prosecutors can actually prove the export paperwork was knowingly and intentionally fraudulent That is the hard part It is They must definitively trace the physical server units from the U S port to their final corporate destinations in China This trial will rigorously test the actual strength and viability of U S export enforcement in the A I era OK we are now moving

into our quick read section First up Anthropic adds Claude Code mods Right On October 1st Anthropic introduced mods to Claude Code OK These are small customizable TypeScript functions They allow developers to dynamically rewrite prompts and completely replace built in platform features For example a developer successfully replaced the built in slash diff feature with a custom version That is useful It is but there is a major security tradeoff to understand These mods completely inherit Claude Code's underlying machine access They do not run in an isolated secure sandbox Next up Stability AI Sean Parker announced a highly anticipated planned update for digital musicians This is cool Stability

AI will soon let users directly guide audio generation by humming or beatboxing into the microphone There is no official release date yet The company recently secured a massive 76 million funding round That new funding features fully licensed music catalogs from Sony Warner and Universal Music Group Meanwhile Cloudflare releases open models for AI decisions without text generation The new models are called Clef and Clef Flash They make AI decisions directly using raw mathematical probabilities So no text Right They entirely skip the slow process of text generation They claim lightning fast median processing latencies Clef hits an incredible 38 8 milliseconds That is fast For comparison the

older Jeff model takes 524 1 milliseconds Cloudflare does warn that model accuracy varies sharply depending on the specific task Finally OpenAI schedules three GPT 5 API models for shutdown They sent a formal developer notice on October 1st Right GPT 5 3 Codex GPT 5 1 and GPT 5 4 Nano will permanently shut down on April 1st 2027 Good to know Developers have exactly six months to migrate their existing applications OpenAI highly recommends migrating to newer replacements like GPT 6 Sol and Luna We are now moving to our three takeaways of the week First the definition and regulation of AI are evolving rapidly This spans from

government rebrands creating domain gold rushes to models stripping out text entirely to make probability based decisions Second the barrier to enterprise AI is no longer the intelligence of the models but the human infrastructure Companies are realizing they must heavily invest in training human engineers to bridge the gap between AI capabilities and actual business deployment Third as AI interaction crosses the uncanny valley into real time highly persuasive video avatars the pressure on physical hardware supply chains is sparking black markets and international smuggling rings Next week you should watch the industry's evolving response to AI disclosure and safety protocols This is particularly urgent as systems like Tavis

blur the line between human and machine interaction Before we go consider this If models are already scoring 100 percent on isolated accounting tests and Anthropic is building an army of humans to integrate those models how long until the integration engineers are the only humans left in the corporate finance department You can find more coverage at superpowerdaily com Thank you for listening We'll see you next week