Wouldn’t it be cool if you could cut the grass with lasers? Everyone knows that lasers are basically magic, and if you strap a diode laser or two to a lawn mower, it should slice through those pesky blades of grass with zero effort. Cue [Allen Pan]’s video on doing exactly this, demonstrating in the process that we do in fact live in a physics-based universe, and lasers are not magical light sabers that will just slice and dice without effort.

The first attempt to attach two diode lasers in a spinning configuration like the cutting blades on a traditional lawn mower led to the obvious focusing issues (fixed by removing the focusing lenses) and short contact time. Effectively, while these diode lasers can cut blades of grass, you need to give them some time to do the work. Naturally, this meant adding more lasers in a stationary grid, like creating a Resident Evil-style cutting grid, only for grass instead of intruders.

Does this work? Sort of. Especially thick grass has a lot of moisture in it, which the lasers have to boil off before they can do the cutting. As [Allen] and co-conspirator found out, this also risks igniting a lawn fire in especially thick grass. The best attempt to cut the lawn with lasers appears to have been made two years ago by [rctestflight], who used a stationary, 40 watt diode laser sweeping across an area. When placed on a (slowly) moving platform this could cut the lawn in a matter of days, whereas low-tech rapidly spinning blades would need at least a couple of minutes.

Obviously the answer is to toss out those weak diode lasers and get started with kW-level chemical lasers. We’re definitely looking forward to seeing those attempts, and the safety methods required to not turn it into a laser safety PSA.

youtube.com/embed/a0H0vOWUjbY?…

hackaday.com/2024/09/03/mowing…

Once upon a time, machine learning was an arcane field, the preserve of a precious few researchers holed up in grand academic institutions. Progress was slow, and hard won. Today, however, just about anyone with a computer can dive into these topics and develop their own machine learning systems.

Shawn Hymel has been doing just that, in his work in developer relations and as a broader electronics educator. His current interest is reinforcement learning on a tiny scale. He came down to the 2023 Hackaday Supercon to tell us all about his work.

Rewards Are Everything

Shawn finds reinforcement learning highly exciting, particularly when it comes to robotics. “We’re now getting into the idea of, can robots not just do a thing you tell them to, but can they learn to do the thing you tell them to?” he says. Imagine a robot copter that learns to fly itself, or a self-driving car that intuitively learns to avoid pedestrians. The dream is robots that do not simply blindly follow orders, but learn and understand intuitively what to do.
The reinforcement learning system for controlling an inverted pendulum.
Obviously, a great deal of machine learning research involves teams of PhDs and millions of dollars in funding. As an individual, Shawn decided to start smaller. Rather than try and build an advanced quadripedal robot that could teach itself to walk, he instead started with a simple inverted pendulum. It’s a classical control theory project, but he set about getting it to work with reinforcement learning instead.

Reinforcement learning is all about observation. The AI in charge of the inverted pendulum can see the position of the pendulum, and its angular velocity. It can also swing it around with a stepper motor, and knows the stepper motor’s angle and velocity. The reinforcement part involves setting a “reward” for the desired position of the pendulum—namely, when it’s balancing in the inverted position. Thus, over time, the AI learns which actions correspond to this reward, and it effectively “learns” how to control the system.

As is often the way, Shawn’s first attempts didn’t work. There was too much latency between the measurements from the inverted pendulum being sent from an Arduino via serial reaching the AI agent running on his computer’s GPU. Thus, he simplified things. Instead of trying to get the pendulum to balance, he decided to just try and teach an AI to swing it up vertically. He also decided to run the AI on a microcontroller, eliminating much of the latency involved in trying to train a model on his GPU. He also simplified the action space—rather than continuous control of the stepper, the AI was only able to make three actions. Either add 10 degrees, subtract 10 degrees, or do nothing.
The initial reward function is developed to optimize keeping the pendulum vertical and not moving.
This proved far more successful. He was able to train a model using the Stable Baselines3 framework that could successfully make positive actions towards flipping up the pendulum. Once trained using the actor-critic method, the actor half of the model could be deployed to a microcontroller and tested on the real system.

He used Edge Impulse to compress the model and ran it on a Seeed Studio XIAO ESP32S3. The microcontroller no longer needs to run the reward function, as it’s already been trained on how to act to get the desired result. It just goes ahead and does its thing. The live demo worked, too — the model was able to swing the pendulum (briefly) into the vertical position.

Finding the Best Tool for the Job

Shawn notes that for reinforcement learning tasks like these, virtual training grounds can be of great value. They allow much training to happen much faster, often with thousands of iterations running in parallel. There’s also less hassle versus training models on real mechanical hardware, which can get damaged or require manual resets after each training run.

All this is not to say that reinforcement learning is the be-all and end-all of robotics these days. As Shawn explains, for many tasks, particularly straightforward and repetitive ones, classical control theory remains supreme. Just because you can do a task with machine learning techniques, doesn’t mean it’s the best way to go about it.

Ultimately, reinforcement learning can help a machine achieve all kinds of complicated tasks. The trick is to create the right reward function and measure the right parameters. As Shawn ably demonstrates, choosing an appropriately simple goal is also a great way to get started!

youtube.com/embed/50R5EsQB5Rw?…

hackaday.com/2024/09/03/superc…

Nvidia ha utilizzato video da YouTube e altre fonti per addestrare i suoi prodotti AI, secondo le e-mail interne e i documenti ottenuti da 404 Media.

Discutendo gli aspetti legali ed etici dell’utilizzo di contenuti protetti da copyright per addestrare modelli di intelligenza artificiale, Nvidia ha affermato che le loro azioni sono pienamente conformi alla legge sul copyright. Conversazioni interne tra i dipendenti di Nvidia mostrano che quando i dipendenti sollevavano domande su potenziali questioni legali, i manager assicuravano loro che hanno il permesso di utilizzare i dati da parte del top management dell’azienda.

Un ex dipendente di Nvidia ha affermato che ai lavoratori è stato chiesto di scaricare video da Netflix, YouTube e altre fonti per addestrare modelli di intelligenza artificiale, come il generatore di mondi 3D Omniverse, sistemi di auto a guida autonoma e prodotti come “umani digitali”. Il progetto, denominato Cosmos, non è stato ancora presentato al grande pubblico.

L’obiettivo di Cosmos è creare un modello di generazione video all’avanguardia in grado di simulare luce, fisica e intelligenza in un unico posto, permettendo a Cosmos di essere usato in una varietà di applicazioni. Dalle comunicazioni interne emerge che i dipendenti hanno utilizzato il programma open source yt-dlp per scaricare video da YouTube, aggirando i blocchi attraverso macchine virtuali con indirizzi IP aggiornati.

I project manager hanno parlato dell’utilizzo di 20-30 macchine virtuali in Amazon Web Services utilizzate per scaricare l’equivalente di 80 anni di video ogni giorno. A maggio, un portavoce di Nvidia ha detto che la società stava completando la prima versione della sua pipeline di dati e si stava preparando a costruire una fabbrica di dati video che avrebbe “generato ogni giorno una quantità di dati pari a una vita intera“.

Google e Netflix hanno confermato che l’utilizzo dei loro contenuti da parte di Nvidia costituisce una violazione dei loro termini di servizio. Ai dipendenti di Nvidia con problemi legali è stato detto dai manager che si trattava di una “decisione esecutiva” e che non dovevano preoccuparsene.

Tuttavia, molti ricercatori e avvocati sostengono che l’uso di contenuti protetti da copyright per addestrare l’intelligenza artificiale sia una questione legale aperta. Negli ultimi anni è diventato più comune per gli accademici concedere in licenza i propri dati di ricerca per uso non commerciale.

Il progetto Cosmos prevedeva l’uso di video sia pubblici che interni, nonché di dati raccolti dai ricercatori. Tuttavia, le licenze per molti di questi set di dati ne limitano l’uso ai soli scopi accademici

L'articolo Nvidia sotto accusa: video YouTube e Netflix usati per addestrare AI senza permesso proviene da il blog della sicurezza informatica.

!Programmer Humor
Hi, this is a question that popped into my mind when i saw an article about some AWS engineer talking about ai assistants taking over the job of programmers, this reminded me that it's not the first time that something like this was said.

My software engineering teacher once told me that a few years ago people believed graphical tools like enterprise architect would make it so that a single engineer could just draw a pretty UML diagram and generate 90% of the project without touching any code,
And further back COBOL was supposed to replace programmers by letting accountants write their own programs.

Now i'm curious, were there many other technologies that were supposedly going to replace programmers that you remember?

i hope someone that's been around much more than me knows something more or has some funny stories to share

Once the nerve center of Windows operating systems, the Control Panel and its multitude of applets has its roots in the earliest versions of Windows. From here users could use these configuration applets to control and adjust just about anything in a friendly graphical environment. Despite the lack of any significant criticism from users and with many generations having grown up with its familiar dialogs, it has over the past years been gradually phased out by the monolithic Universal Windows Platform (UWP) based Settings app.

Whereas the Windows control panel features an overview of the various applets – each of which uses Win32 GUI elements like tabs to organize settings – the Settings app is more Web-like, with lots of touch-friendly whitespace, a single navigable menu, kilometers of settings to scroll through and absolutely no way to keep more than one view open at the same time.

Unsurprisingly, this change has not been met with a lot of enthusiasm by the average Windows user, and with Microsoft now officially recommending users migrate over to the Settings app, it seems that before long we may have to say farewell to what used to be an intrinsic part of the Windows operating system since its first iterations. Yet bizarrely, much of the Control Panel functionality doesn’t exist yet in the Settings app, and it remain an open question how much of it can be translated into the Settings app user experience (UX) paradigm at all.

Considering how unusual this kind of control panel used to be beyond quaint touch-centric platforms like Android and iOS, what is Microsoft’s goal here? Have discovered a UX secret that has eluded every other OS developer?

A Simple Concept

The Windows 3.1 Control Panel (1992). (Source: ToastyTech.com)
Settings which a user may want to tweak on their computer system range from hardware devices and networks to the display resolution and wallpaper, so it makes sense to put all of these configuration options within an easy to reach and use location. Generally this has meant something akin to a folder containing various clickable icons and accompanying text which together make clear what settings can be configured by opening it. In addition, the same setting dialogs can be accessed using context-sensitive menus, such as when right-clicking on the desktop.
The Windows 98 Control Panel. (Source: ToastyTech.com)
It’s little wonder that for the longest time operating systems have settled for this approach, as it is intuitive, and individual items can have stylized icons that make it even more obvious what settings can be configured by clicking on it, such a keyboard, a mouse, a display, etc. As graphical fidelity increased, so did the styling of these icons, with MacOS, Windows, BeOS and the various desktop environments for OSs like the Linuxes and BSDs all developing their own highly skeuomorphic styles to make their UIs more intuitive and also more pleasant to look at. A good overview of the Windows Control Panel evolution can be found over at the Version Museum website.
The Windows XP Control Panel in ‘Classic’ view. (2001) (Source: suffieldacademy.org)
Coming from the still somewhat subdued style of Windows XP after years of Windows 9x and Windows NT/2000, Windows Vista and Windows 7 cranked this style up to eleven with the Windows Aero design language. This meant glass, color, translucency, depth and high-fidelity icons that made the function of the Control Panel’s individual entries more obvious than ever, creating a masterpiece that would be very hard to beat. The user was also given two different ways to view the Control Panel: the simplified category-based view, or the ‘classic’ view with all icons (and folders for e.g. Administrative Tools) visible in one view.
Windows 7 Control Panel (2009) in category view. (Source: techrepublic.com)
Meanwhile Apple did much the same thing, leaning heavily into their unique design language not only for its desktop, but ultimately also for its mobile offerings. Everything was pseudo-3D, with vivid colors adorning detailed renderings of various physical items and so on, creating a true feast for the eyes when taking in these lush UIs, with efficient access to settings via clearly marked tabs and similar UI elements.
The Mac OS X Panther System Preferences in 2003. (Source: Gadget Unity TV)
This way of organizing system settings was effectively replicated across a multitude of environments, with operating systems like Haiku (based on BeOS) and ReactOS (re-implementing Windows) retaining those classical elements of the original. A truly cross-platform, mostly intuitive experience was created, and Bliss truly came to the computing world.

Naturally, something so good had no right to keep existing, ergo it had to go.

The World Is Flat

The first to make the big change was Microsoft, with the release of Windows 8 and its Metro design language. This new visual style relied on simple shapes, with little to no adornments or distractions (i.e. more than a single color). Initially Microsoft also reckoned that Windows users wanted every window to be full-screen, and that hot edges and sides rather than a task bar and start menu was the way to go, as every single system running Windows 8 would obviously have a touch screen. Fortunately they did backtrack on this, but their attempt to redesign the Control Panel into something more Metro-like with the Settings app did persist, like an odd growth somewhere on a body part.
Windows 8’s PC Settings app (2012). (Source: softpedia.com)
Although the Control Panel remained in Windows 8 as well, the course had been set. Over time this small lump developed into the Settings app in Windows 10, by which time Metro had been renamed into the Microsoft Design Language (MDL), which got a recent tweak in what is now called the Fluent Design Language (FDL) for Windows 11.

Central to this is the removal of almost all colors, the use of text labels over icons where possible (though simple monochrome icons are okay) and only rectangles with no decorations. This also meant no folder-centric model for settings but rather all the items put into a text-based menu on the left-hand side and an endless scroll-of-doom on the right side containing sparsely distributed settings.

This led to the absolutely beautifully dystopian Settings app as it exists in Windows 10:
The Settings app in Windows 10 back in ~2015. Hope you don’t like colors.
All of this came as skeuomorphic designs were suddenly considered ‘passé’, and the new hotness was so-called Flat Design. Google’s Material Design as developed in 2014 is another good example of this, with the characteristic ‘flat UI elements adrift in a void’ aesthetic that has now been adopted by Microsoft, and a few years ago by Apple as well starting in 2022 with MacOS Ventura’s System Settings (replacing System Preferences).
Monterey’s General system preferences (left) are different from Ventura’s General system settings (right). (Credit: MacWorld)
Rather than a tabbed interface to provide a clear overview, everything is now a blind hierarchy of menu items to scroll through and activate to access sub-, sub-sub-, and sub-sub-sub- items, and inevitably realize a few times that you’re in the wrong section. But rather than being able to click that other, correct tab, you now get to navigate back multiple views, one click at a time.

It isn’t just Windows and Apple either, but many of the big desktop environments like Gnome have also moved to this Flat Design Language. While various reasons have been provided for these changes, it’s undeniable that FDL makes a UI less intuitive (because there’s less useful visual information) and makes for a worse user experience (UX) with worse ergonomics as a result (because of the extra scrolling and clicking). This is especially obvious in the ‘independent applets’ versus ‘monolithic settings app’ comparison.

One-Track Mind

Imagine that you’re trying out a couple new wallpapers in Windows while keeping an eye on Windows Update’s latest shenanigans. You then need to quickly adjust the default audio device or another small adjustment unrelated to any of these other tasks. If you are using Windows 7 or earlier with the Control Panel applets, this is normal behavior and exceedingly common especially during hardware troubleshooting sessions.

If you’re using the Settings app, this is impossible, as only view can be active at a given time. You think you’re smart and right-click the desktop for ‘Personalize desktop’ so that the other Settings view stays intact? This is not how it works, as the Settings app is monolithic and now shifts to the newly selected view. Currently this is not too noticeable yet as many applets still exist in Windows 10 and 11, but as more and more of these are assimilated into the Settings app, such events will become more and more common.

It would seem that after decades of UI and UX evolution, we have now reached a definite point where UX is only getting worse, arguably around the release of Windows 8. With color banished, anything even remotely pseudo-3D frowned upon and UIs based around touch interfaces, there will soon be no difference between using a desktop PC, tablet or smartphone. Just in the worst way possible, as nobody has ever written about the amazing ergonomics and efficient UX of the latter two devices.

Perhaps our only hope may lie with the OSes and desktop environments that keep things real and stick to decades of proven UX design rather than give into Fad Driven Development.

Rest in peace, Windows Control Panel. We hope to see you again soon in ReactOS.

hackaday.com/2024/09/03/a-wind…

I ricercatori di sicurezza hanno scoperto otto nuove vulnerabilità nelle versioni macOS delle applicazioni Microsoft (Outlook, Teams, Word, Excel, PowerPoint e OneNote), che, se sfruttate, consentono agli aggressori di aumentare i diritti e ottenere l’accesso a dati riservati.

Secondo una descrizione di Cisco Talos, i difetti identificati aiutano a bypassare le impostazioni sulla privacy nel sistema operativo, che sono basate sul framework TCC (Trasparenza, Consenso e Controllo).

“Se un aggressore sfrutta le vulnerabilità scoperte, potrà ottenere tutti i diritti concessi da Microsoft sulle applicazioni interessate”, scrivono gli esperti .

“Ad esempio, un utente malintenzionato potrebbe inviare e-mail dall’account della vittima, nonché registrare audio e video, senza alcuna interazione con l’utente preso di mira.” hanno aggiunto.

In teoria, un utente malintenzionato può inserire librerie dannose in una qualsiasi di queste applicazioni, cosa che gli permetterà non solo di ottenere i diritti di quest’ultima, ma anche di estrarre una serie di dati riservati. Come notano gli esperti, per uno sfruttamento efficace l’aggressore deve già avere un certo accesso al sistema della vittima.

TCC e protezione dei dati

TCC applica una politica che richiede alle applicazioni di ottenere il consenso esplicito dell’utente prima di poter accedere a risorse protette quali contatti, calendari, foto e posizione, garantendo che gli utenti mantengano il controllo diretto sulle proprie informazioni personali e sull’accessibilità dei propri dati alle applicazioni.

TCC opera congiuntamente agli entitlement, che sono un set di capacità richieste per la funzionalità di un’app. Gli sviluppatori scelgono questi entitlement da una selezione fornita da Apple e, sebbene solo un sottoinsieme di tutti i possibili entitlement sia disponibile per l’uso generale, i più potenti sono riservati esclusivamente alle applicazioni e ai binari di sistema di Apple.

Quando un’applicazione con diritti specifici richiede inizialmente l’utilizzo di una specifica funzionalità, viene visualizzato un pop-up di autorizzazione.

L’immagine sopra mostra un esempio di tale richiesta di autorizzazione: “Malevolent App” vorrebbe accedere alla telecamera. L’utente deve decidere se consentire o negare l’accesso alla telecamera. Questa decisione viene quindi registrata nel database TCC.

Una volta che l’utente ha fatto la sua scelta, qualsiasi futura richiesta relativa alla telecamera da parte della “Malevolent App” sarà regolata dalla decisione registrata nel database. Questo sistema consente effettivamente agli utenti di controllare ed essere informati delle azioni sensibili alla privacy che un’applicazione intende eseguire. La necessaria interazione dell’utente è ciò che consente agli utenti di impedire alle applicazioni dannose di eseguire azioni sensibili come la registrazione di un video o lo scatto di foto.

Un utente può successivamente verificare questa autorizzazione nella sezione “Privacy e sicurezza” delle “Impostazioni di sistema” di macOS. Lì, è possibile trovare un elenco di autorizzazioni, tra cui Fotocamera, Microfono e Servizi di localizzazione.

L'articolo Le App Microsoft su macOS possono essere utilizzate per accedere ai dati riservati dell’Utente proviene da il blog della sicurezza informatica.

“Ogni mio nuovo lavoro rappresenta la summa e la sintesi di tutti quelli realizzati prima, evolvendone nuovamente il significato e la portata artistica. Credo che “Strategia Esoterica” abbia comunque una forza molto superiore perché frutto di una trasmutazione molto potente”- Deca @Musica Agorà

iyezine.com/deca-strategia-eso…

Deca - Strategia esoterica

Deca - Strategia esoterica - “Ogni mio nuovo lavoro rappresenta la summa e la sintesi di tutti quelli realizzati prima, evolvendone nuovamente il significato e la portata artistica.

^{Massimo Argo (In Your Eyes ezine)}