La Cina supera gli Stati Uniti nei modelli di intelligenza artificiale open source

La Cina supera gli Stati Uniti nel mercato globale dei modelli di intelligenza artificiale (IA) open source. Secondo un rapporto del Financial Times del 26 novembre, uno studio del MIT e della startup di IA open source Hugging Face ha rilevato che nell’ultimo anno la quota di download di modelli di IA open source sviluppati da team cinesi è salita al 17%, superando per la prima volta le controparti americane e ottenendo un vantaggio fondamentale nell’applicazione globale della tecnologia IA.

Nel contesto del boom dell’intelligenza artificiale che sta travolgendo l’industria tecnologica globale, i giganti tecnologici statunitensi come OpenAI, Google e Anthropic tendono ad adottare una strategia “chiusa“, mantenendo il controllo completo sulle tecnologie di intelligenza artificiale avanzate e traendo profitto dagli abbonamenti degli utenti e dalle partnership aziendali.

Al contrario, le aziende tecnologiche cinesi privilegiano una strategia più aperta, rilasciando una serie di modelli open source.
Screenshot di un rapporto del Financial Times che mostra la percentuale di download di modelli open source misurata da Hugging Face.
I modelli di intelligenza artificiale open source si riferiscono a modelli di intelligenza artificiale il cui codice e architettura sono pubblicamente disponibili, consentendo a qualsiasi sviluppatore di scaricarli, utilizzarli, modificarli e distribuirli. L’adozione diffusa di modelli open source avrà un impatto significativo sullo sviluppo futuro dell’intelligenza artificiale, facilitando lo sviluppo di prodotti da parte delle startup e il miglioramento dei ricercatori.

Grazie all’impegno instancabile dei ricercatori cinesi, i modelli open source cinesi hanno dimostrato una forte competitività. Una ricerca del MIT e di Hugging Face ha rilevato che nell’ultimo anno la percentuale di nuovi modelli di intelligenza artificiale open source sul totale dei download di modelli open source è salita al 17%, superando il 15,8% dei team di sviluppo negli Stati Uniti. Questa è la prima volta che i team cinesi hanno superato le loro controparti americane in questo parametro.

Wendy Chang, analista senior del Mercator Institute for China Studies, un think tank tedesco, sottolinea: “L’open source è una tendenza più diffusa in Cina che negli Stati Uniti. Le aziende americane non sono disposte a farlo; vogliono fare soldi e non vogliono rivelare i loro segreti commerciali”.

Secondo i dati del MIT e di Hugging Face, DeepSeek e Qwen di Alibaba Cloud sono i modelli open source più scaricati dalla Cina. Il Financial Times ha riportato che il modello DeepSeek-R1 ha scioccato la Silicon Valley: utilizza costi e potenza di calcolo bassi, ma le sue prestazioni sono paragonabili ai migliori modelli statunitensi, sollevando dubbi sulla capacità dei laboratori di intelligenza artificiale statunitensi di mantenere il loro vantaggio competitivo.

Esperti del settore hanno dichiarato al Financial Times che, nonostante la serie di misure adottate dagli Stati Uniti per reprimere la Cina, come i controlli sulle esportazioni di chip, la Cina ha un gran numero di talenti eccezionali che hanno dimostrato grande creatività nello sviluppo di modelli open source.

Nel tentativo di vincere la cosiddetta “corsa all’intelligenza artificiale”, l’amministrazione Trump stava cercando di convincere le aziende americane a investire in modelli open source che incarnassero i “valori americani”. Tuttavia, il numero di grandi team di sviluppo indipendenti per modelli open source negli Stati Uniti è molto inferiore rispetto alla Cina, e giganti della tecnologia statunitensi come Meta stanno ora scegliendo di aumentare i loro investimenti in modelli closed source.

Ad agosto di quest’anno, OpenAI ha rilasciato il suo primo lotto di modelli “open weights”. Questi modelli sono gratuiti, ma le informazioni che forniscono non sono complete come quelle dei modelli open source e non includono il codice e i dati di training necessari per addestrare un modello da zero.

L'articolo La Cina supera gli Stati Uniti nei modelli di intelligenza artificiale open source proviene da Red Hot Cyber.

Any Old TV Can Be A Clock With Arduino

If you’ve got an old black and white TV, it’s probably not useful for much. There are precious few analog broadcasters left in the world and black and white isn’t that fun to watch, anyway. However, with a little work, you could repurpose that old tube as a clock, as [mircemk] demonstrates.

The build is based around an Arduino Nano R3. This isn’t a particularly powerful microcontroller board, but it’s good enough to run the classic TVOut library. This library lets you generate composite video on an Atmel AVR microcontroller with an absolute minimum of supporting circuitry. [mircemk] paired the Arduino with a DS3231 real-time clock, and whipped up code to display the time and date on the composite video output. He then also demonstrates how to hack the signal into an old TV that doesn’t have a specific input for composite signals.

You’ll note the headline says “any old TV can be a clock,” and that’s for good reason. Newer TVs tend to eschew the classic composite video input, so the TVOut library won’t be any good if you’re trying to get a display up on your modern-era flatscreen. In any case, we’ve seen the TVOut library put to good use before, too. Video after the break.

youtube.com/embed/JuSsQqInKgo?…

hackaday.com/2025/11/27/any-ol…

Today #Putin announced that #Russia is “conducting military operations in Komsomolsk” and even “took a number of buildings”.

Most Russian news portals simply repeated his ramblings as usual without much checking, but Lenta tried to be helpful and explained that Komsomolsk is the Soviet name of Horishni Plavni, a town in Poltava oblast’.[^1]

The problem is that Horishni Plavni is located in central Ukraine, around ~250 km from nearest active combat zone! But even if Lenta made a mistake, the Komsomolsk still doesn’t exist - any nearest location with similar name (Komsomolskyi) is located in Luhansk oblast[^2] and occupied since 2014. There’s also two Komsomolsk in the Crimea.

What Putin had in mind will probably forever remain a mystery, but it clearly demonstrates the level of “journalism” in Russia AD 2025 - they’re brainlessly repeating whatever Kremlin reports with zero verification or questions.

[^1]: lenta.ru/news/2025/11/27/putin…

[^2]: deepstatemap.live/#dl!coords!4…

Boosting Antihydrogen Production using Beryllium Ions

Antihydrogen forms an ideal study subject for deciphering the secrets of fundamental physics due to it being the most simple anti-matter atom. However, keeping it from casually annihilating itself along with some matter hasn’t gotten much easier since it was first produced in 1995. Recently ALPHA researchers at CERN’s Antimatter Factory announced that they managed to produce and trap no fewer than 15,000 antihydrogen atoms in less than seven hours using a new beryllium-enhanced trap. This is an eight-fold increase compared to previous methods.

To produce an antihydrogen atom from a positron and an antiproton, the components and resulting atoms can not simply be trapped in an electromagnetic field, but requires that they are cooled to the point where they’re effectively stationary. This also makes adding more than one of such atom to a trap into a tedious process since the first successful capture in 2017.

In the open access paper in Nature Communications by [R. Akbari] et al. the process is described, starting with the merging of anti-protons from the CERN Antiproton Decelerator with positrons sourced from the radioactive decay of sodium-22 (β⁺ decay). The typical Penning-Malmberg trap is used, but laser-cooled beryllium ions (Be⁺) are added to provide sympathetic cooling during the synthesis step.

Together with an increased availability of positrons, the eight-fold increase in antihydrogen production was thus achieved. The researchers speculate that the sympathetic cooling is more efficient at keeping a constant temperature than alternative cooling methods, which allows for the increased rate of production.

hackaday.com/2025/11/25/boosti…

US Patent Changes Promise Severe Consequences

When someone creates a US patent, they go through a review process to stop the most blatant copies from previous patents or pre-existing work. After this, you may still have bad patents get through, which can be removed through litigation or publicly accessible methods such as Inter Partes Review (IPR). The latter of which is planned to be changed as we know it in the near future.

IPR is a method where an individual can claim that an existing patent is invalid due to pre-existing work, such as something the individual should have creative ownership over. While there is always the litigation method of removing blatantly fraudulent patents, a small business or the average person is unlikely to have the funds.

New regulations are changing how IPRs can be filed in some substantial ways. Now, if someone files an IPR, they give up the right to future litigation on their rights over a patent. This is obviously not ideal for someone who may have their own products on the line if an IPR is to fail. Additionally, IPRs will no longer be able to be even tried if there are existing cases against the patent, even under poor previous cases. While this change is meant to increase the efficiency of the patent office, there are some serious consequences that must be looked into either way. The patent office also cites IPRs being beneficial to larger organizations rather than the smaller businesses, though you can make your own conclusions based on the U.S. Patent and Trademark Office’s arguments here.

Hackaday certainly can not give any legal advice on how this change will affect you, but there are cases given by both sides that may persuade you to write to your legal representatives if you live in the States. Even still, we here at Hackaday have seen our fair share of patent trolls causing issues. If you want a case of blatant patent shenanigans check out these 3D printing layers that promise improved strength!

Thanks [patentTrollsAreTheWorst] for the tip!

hackaday.com/2025/11/27/us-pat…

Designing PLA to Hold Over a Metric Ton

There’s never been such a thing as being “too competitive” when it comes to competition. This is something that [Tom Stanton] from Tim Stanton (wait, what now), [Tom]’s 2nd channel, took to heart for Polymaker’s 3D design challenge. The goal was simple: a single 3D printed part to hold as much weight as possible.

While seemingly simple, when considering the requirements, including a single print in addition to being able to open up for the mounts, the challenge gets exponentially more complicated. While the simplest and strongest joint would be a simple oval for uniform stress, this isn’t possible when considering the opening requirements. This creates a need for slightly more creativity.

[Tom] starts out with two flat C-shaped geometries to test his design. The design includes teeth specially placed to allow the forces to increase their own strength as force is applied. Flat features have the unfortunate quality of being able to slide across each other rather easily, which was the case during testing; however, the actual structures held up rather well. Moving onto the final design, including a hollow cavity and a much thicker depth, showed good promise early on in the competition, leading up to the finals. In fact, the design won out over anything else, getting over double the max strength of the runner up. Over an entire metric ton, the piece of plastic proved its abilities far past anything us here at Hackaday would expect from a small piece of PLA.

Design can be an absolute rabbit hole when it comes to even the simplest of things, as shown with this competition. [Tom] clearly showed some personal passion for this project; however, if you haven’t had the chance to dive this deep into CADing, keep sure to try out something like TinkerCAD to get your feet wet. TinkerCAD started out simple as can be but has exploded into quite the formidable suite!

youtube.com/embed/GEHNijssAKc?…

hackaday.com/2025/11/27/design…

One-Way Data Extraction For Logging On Airgapped Systems

If you want to protect a system from being hacked, a great way to do that is with an airgap. This term specifically refers to keeping a system off any sort of network or external connection — there is literally air in between it and other systems. Of course, this can be limiting if you want to monitor or export logs from such systems. [Nelop Systems] decided to whip up a simple workaround for this issue, creating a bespoke one-way data extraction method.

The concept is demonstrated with a pair of Raspberry Pi computers. One is hooked up to critical industrial control systems, and is airgapped to protect it against outside intruders. It’s fitted with an optocoupler, with a UART hooked up to the LED side of the device. The other side of the optocoupler is hooked up to another Raspberry Pi, which is itself on a network and handles monitoring and logging duties.

This method creates a reliable one-way transmission method from the airgapped machine to the outside world, without allowing data to flow in the other direction. Indeed, there is no direct electrical connection at all, since the data is passing through the optocoupler, which provides isolation between the two computers. Security aficionados will argue that the machine is no longer really airgapped because there is some connection between it and the outside world. Regardless, it would be hard to gain any sort of access through the one-way optocoupler connection. If you can conceive of a way that would work, drop it down in the comments.

Optocouplers are very useful things; we’ve seen them used and abused for all sorts of different applications. If you’ve found some nifty use for these simple parts, be sure to drop us a line!

hackaday.com/2025/11/27/one-wa…

Hydrofoil Bikes Are Harder To Build Than You Think

Hydrofoils are perhaps best known for their application on boring ferries and scary boats that go too fast. However, as [RCLifeOn] demonstrates, you can also use them to build fun and quirky personal watercraft. Like a hydrofoil bike! Only, there are some challenges involved.

Hydrofoils work much like airfoils in air. The shape of the foil creates lift, raising the attached vehicle out of the water. This allows the creation of a craft that can travel more quickly because the majority of its body is not subject drag from the water. The key is to design the craft such that the hydrofoils remain at the right angle and depth to keep the craft lifted out of the water while remaining stable.

The hydrofoil bike is created out of a combination of plywood, foam, and 3D printed components. It uses a powerful brushless motor for propulsion, and that’s about it. Sadly, despite the simplicity, it wasn’t an instant success. As you might expect, balancing on the bike is quite difficult, particularly when trying to get it started—as the foils need some speed to actually start generating meaningful lift.

After further research into commercial hydrofoil bikes, [RCLifeOn] realized that the buoyancy of the bike made it too hard to straddle when starting out. Some of the 3D printed foils also proved more than a little fragile. It’s back to the drawing board for now—the power system is likely up to snuff, but the dynamics of the platform need work. It’s perhaps no surprise; we’ve covered the challenges of hydrofoil stability before. If you want to go fast on water, you could go the easier route and just build an electric surfboard. Video after the break.

youtube.com/embed/zP1nS3sIu2U?…

hackaday.com/2025/11/27/hydrof…

La Truffa del CEO! l’inganno che sta travolgendo le aziende italiane

Questa mattina Paragon Sec è stata contattata da un’azienda italiana vittima di un nuovo tentativo di frode conosciuto come Truffa del CEO. L’ufficio contabilità ha ricevuto un’e-mail urgente, apparentemente inviata dal loro Amministratore Delegato, contenente la richiesta di effettuare con immediatezza il pagamento di una fattura da 4.000 euro.

Il messaggio, accompagnato da una fattura apparentemente autentica, indicava la necessità di un bonifico immediato. Il dipendente incaricato dei pagamenti, convinto di eseguire un ordine diretto del proprio dirigente, ha effettuato il trasferimento senza ulteriori verifiche.

Solo successivamente la banca ha rilevato che l’IBAN indicato era associato a un soggetto fraudolento e ha bloccato l’operazione in tempo, impedendo la perdita economica. Si tratta di un caso che conferma come queste campagne stiano diventando sempre più frequenti, raffinate e mirate alle aziende italiane.
Documento fatto circolare all’interno dell’email truffa (fonte Paragon Sec)

Che cos’è la Truffa del CEO

La Truffa del CEO, conosciuta a livello internazionale come Business Email Compromise, è una tecnica di ingegneria sociale in cui i criminali si spacciano per un alto dirigente dell’azienda, tipicamente l’Amministratore Delegato o il Direttore Finanziario.

Utilizzando e-mail costruite in modo credibile, i truffatori inducono un dipendente fidato – spesso chi gestisce i pagamenti – a eseguire trasferimenti di denaro urgenti e apparentemente legittimi.

Il punto di forza di questo attacco non è la tecnologia, ma la manipolazione psicologica: urgenza, autorevolezza e riservatezza vengono sfruttate per spingere la vittima ad agire senza riflettere.
Email inviata ad un responsabile acquisti di una azienda contenente lafattura truffa (fonte Paragon Sec)

Come difendersi

Per contrastare la Truffa del CEO, è fondamentale adottare procedure interne chiare e formare i dipendenti a riconoscere segnali sospetti.
Tra le misure più efficaci rientrano:

• Verificare sempre con attenzione l’indirizzo e-mail del mittente.
• Prestare attenzione a cambi di stile comunicativo, errori o richieste insolite.
• Diffidare di messaggi che richiedono segretezza, urgenza o scavalcano le procedure standard.
• Contattare direttamente il dirigente coinvolto tramite un canale alternativo per confermare la richiesta.

Cosa ci insegna questo episodio

Questo caso dimostra come gli attacchi non colpiscano solo la tecnologia, ma soprattutto i comportamenti umani. La vulnerabilità principale risiede nella fiducia, nella pressione psicologica e nella mancanza di una verifica incrociata.

La prevenzione passa attraverso la formazione continua, la consapevolezza e l’adozione di processi aziendali che permettano ai dipendenti di fermarsi, dubitare e verificare prima di eseguire qualunque operazione finanziaria fuori dall’ordinario.

La Truffa del CEO, ancora una volta, si conferma una delle minacce più insidiose per le aziende italiane.

Come si svolge la Truffa del CEO

La Truffa del CEO inizia con una fase di raccolta di informazioni, che i criminali svolgono attraverso tecniche OSINT e web scraping di piattaforme come LinkedIn. Qui ricostruiscono l’organigramma aziendale, identificano il CEO, il CFO e le figure chiave del reparto amministrativo, osservano abitudini, ruoli e relazioni interne. Parallelamente analizzano anche dati provenienti da vecchie collection del dark web, che contengono indirizzi e-mail, conversazioni compromesse e modelli di naming utili a imitare fedelmente la comunicazione interna dell’azienda.

Una volta ottenute queste informazioni, gli attaccanti isolano le due figure centrali del loro schema: il dirigente da impersonare e il dipendente più esposto, solitamente chi si occupa di bonifici o pagamenti. Attraverso social network, archivi pubblici e dati trapelati da precedenti violazioni, ricostruiscono procedure, orari, responsabilità e dettagli personali. Questo consente loro di capire quando il dirigente potrebbe non essere raggiungibile e in quali condizioni il dipendente sarebbe più incline a eseguire un ordine urgente senza verifiche.

Nella fase finale gli attaccanti costruiscono e inviano l’e-mail fraudolenta, sfruttando il linguaggio, la firma e lo stile del dirigente reale. La comunicazione contiene una richiesta urgente di pagamento, spesso accompagnata da termini come “riservato”, “non discutere con altri” o “deve essere fatto subito”.

A quel punto il successo della truffa non dipende più dalla tecnologia, ma dalla pressione psicologica esercitata sulla vittima, che crede di eseguire un ordine legittimo proveniente dall’alto.

L'articolo La Truffa del CEO! l’inganno che sta travolgendo le aziende italiane proviene da Red Hot Cyber.

Google is Building a New OS

Windows, macOS, and Linux are the three major desktop OSs in today’s world. However, there could soon be a new contender, with Google stepping up to the plate (via The Verge).

You’ve probably used Google’s operating systems before. Android holds a dominant market share in the smartphone space, and ChromeOS is readily available on a large range of notebooks intended for lightweight tasks. Going forward, it appears Google aims to leverage its experience with these products and merge them into something new under the working title of “Aluminium OS.”

The news comes to us via a job listing, which sought a Senior Product Manager to work on a “new Aluminium, Android-based, operating system.” The hint is in the name—with speculation that the -ium part of Aluminium indicates its relationship to Chromium, the open-source version of Chrome. The listing also indicated that the new OS would have “Artificial Intelligence (AI) at the core.” At this stage, it appears Google will target everything from cheaper entry level hardware to mid-market and premium machines.

It’s early days yet, and there’s no word as to when Google might speak more officiously on the topic of its new operating system. It’s a big move from one of the largest tech companies out there. Even still, it will be a tall order for Google to knock off the stalwart offerings from Microsoft and Apple in any meaningful way. Meanwhile, if you’ve got secret knowledge of the project and they forget to make you sign an NDA, don’t hesitate to reach out!

hackaday.com/2025/11/27/google…

Stack n’ Rack Your Hardware With the HomeRacker Project

Things are cooler when rack-mounted, and [KellerLab] aims to make that all far more accessible with the HomeRacker, a modular and 3D-printable rack building system designed to let you rack-mount to your heart’s content. While it can handle big things, it seems especially applicable to tasks like mounting one’s home network equipment and Raspberry Pi machines.
A rack is a great place for those Raspberry Pi servers and home networking equipment, but it can also handle bigger jobs.
The basic system (or core) consists of three different parts: supports, connectors, and lock pins. The supports are the main structural bars, the connectors mostly go at the corners, and the lock pins ensure everything stays put. The nominal sizing is a 15 mm x 15 mm profile for the supports, with lengths being a multiple of 15 mm.

All is designed with 3D printing in mind, and requires no tools to assemble or disassemble. There are design elements we really appreciate, like how parts are printed at an angle, which improves strength while eliminating the need for supports. The lock pins (and the slots into which they go) are designed so that they are effective and will neither rattle nor fall out.

But the core system is just the foundation. There’s plenty of modularity and expansions to handle whatever one may need, from Gridfinity shelves and drawers to various faceplates and other modules. There are some example applications available from [KellerLab]’s HomeRacker models page, like CD shelf, under-desk drawer, or filament rack.

[KellerLab] welcomes any collaboration, so check out the GitHub repository for CAD references and design files.

One last point to make about the value of printing objects like this at an angle: not can the resulting layer lines provide better strength and reduce or eliminate the need for supports, but printing at an angle can help hide layer lines.

youtube.com/embed/g8k6X_axYug?…

hackaday.com/2025/11/27/stack-…