Is Your ATSC 3.0 Box Spying On You?

For the better part of the last four years, I have been tracking the rollout of the ATSC 3.0 broadcast standard. While marketed as the future of television, this transition has become a point of significant friction between broadcasters and viewers. The primary source of this tension is the decision by major networks to encrypt over-the-air signals, which frequently requires an active internet connection to watch what has historically been a free, offline medium.

In my latest video, I did some deep packet analysis of two of the most popular ATSC 3.0 TV boxes to see what these boxes are doing while users are watching over the air television.

I set up GMKtec Mini PC with two ethernet ports that I configured as a router to sit between the television boxes and the Internet. This setup allowed me to monitor every request sent from the devices to the outside world. I focused my testing on two popular consumer devices: the ADTH box, which is currently one of the most affordable options, and the Zapperbox. While the payloads of the data packets I examined are themselves encrypted, when these packets are triggered along with the destination addresses provide a clear map of which companies are receiving information about my viewing habits.

Starting with the ADTH box, I found that the device is notably “chatty” in its data transmission. Despite marketing claims that it can function without a constant connection, the box failed to tune into encrypted channels in my tests until it was reconnected to the Internet after being offline for a few months. When I tuned into a standard ATSC 1.0 channel—the older, unencrypted standard—the box immediately sent telemetry data to Tulka.tv, the company responsible for the device’s software. The chatter increased significantly when I switched to an ATSC 3.0 encrypted channel.

As the encrypted NBC affiliate loaded, the box initiated several sessions with Yotta, a UK-based company that manages encryption certificates for American broadcasters. However, the most notable discovery was the immediate communication with Google Tag Manager when the mandatory “interactive features” loaded up after tuning into the channel.

In a web environment, Google Tag Manager is used to build marketing profiles based on user activity. In this context, the ADTH box was essentially reporting my viewing choice to a marketing platform. The activation of interactive features on the broadcast triggered further data flows to Amazon’s Cloudfront and a service known as Firehose, which is designed for high-volume data collection. This telemetry occurred automatically, with no visible option for me to opt out.

Even unencrypted ATSC 3.0 channels on the ADTH box showed evidence of tracking. When I tuned into my local ABC affiliate, which does not currently use encryption, the box continued to send telemetry to many of the same servers. This suggests that the data collection is not purely a function of the encryption itself, but a broader feature of how these new broadcast stacks are designed to operate.

In comparison, the Zapperbox exhibited a different profile. It still requires an Internet connection to fetch the necessary licenses for encrypted channels, but it appeared to be less communicative than the ADTH box. I did not see the Google Tag Manager requests during my time with the Zapperbox, though this may be because the device does not yet support the interactive overlays that trigger much of that specific traffic. Most of its communication was limited to the Yotta license servers and Geniatech, the manufacturer’s technical layer. While it was less “chatty,” it still maintains the requirement of an Internet delivered digital handshake to unlock a broadcast signal.

The transition to ATSC 3.0 represents a fundamental change in the relationship between the broadcaster and the viewer. The anonymity that defined over-the-air television for decades is being replaced by a system that mirrors the tracking found in web browsers and streaming apps. I am looking into ways to further intercept this data to see the specific contents of these packets, but the initial map of activity suggests that the “free” nature of broadcast TV now comes with a different kind of cost beyond the DRM hijacking big broadcasters are after.

See more ATSC 3 analysis here.

SteamOS as a Desktop OS? Browsing, Office, Printing, Local LLM and More!

I have been testing SteamOS on a Minisforum PC equipped with a Ryzen AI 9 370 processor (compensated affiliate link). While much of the attention surrounding this operating system focuses on its console-style interface, there is a functional desktop side that allows for traditional computing tasks.

See the desktop side of SteamOS in my latest video!

The interface is built on KDE Plasma, which offers a layout similar to Windows, including a start menu, taskbar, and file manager. However, the architecture of SteamOS is immutable. This means the operating system is designed to be wiped and replaced during updates. While the user’s files are maintained between updates, system-level modifications, such as kernel drivers or specific VPNs like Tailscale, will not persist after an update without significant technical intervention. This creates a clear boundary between SteamOS and more flexible Linux distributions like Ubuntu or Mint.

For apps, the system relies on the “Discover Store”, which utilizes Flatpaks. These applications run in isolated containers and remain intact through system updates. I found that common applications like Firefox, VLC, and Spotify are readily available. I also installed LibreOffice, which functioned as expected for document and spreadsheet work. While many open-source tools are compatible, high-end professional applications such as DaVinci Resolve are difficult to run due to the specific system architecture and driver requirements.

Another core computing function, printing, is supported on SteamOS. By accessing the system settings, I connected the device to a color laser printer on my network. The installation was straightforward, and the system successfully printed documents from a web browser.

The PC’s 64GB of RAM allowed for an exploration of local AI performance using LM Studio. I loaded a Gemma 4 model to perform image analysis on a photo from my garden. The system identified the objects in the image efficiently, reaching a performance level of 27 tokens per second. While this does not match the speed of a high-end gaming desktop with a dedicated Nvidia GPU, it is a functional level of performance for local AI tasks. It can also serve as a local AI server within a home network, though the aforementioned lack of Tailscale support limits remote access.

For those with game libraries outside of Steam, the Heroic Games Launcher is available through the Discover store. This tool provides access to titles from Epic, GOG, and Amazon Prime. It allows for the configuration of different Proton versions for each game, providing a degree of flexibility for cross-platform libraries.

The transition between the gaming interface and the desktop is handled by a simple toggle. SteamOS is primarily built for gaming appliances, but the desktop environment is sufficient for browsing, office work, and basic media management. And the cool thing is that if you have a Steam Deck, this functionality can essentially turn the Deck into a functional desktop PC when docked.

See my Steam Deck review here!

Take Back Your Android /Google TV Buttons!

If you own an Android or Google TV device, your remote control likely features several branded buttons dedicated to specific streaming services. Manufacturers monetize this hardware real estate by selling priority access to platforms like Netflix or YouTube. For users who do not utilize these specific services, these buttons remain dormant. While hardware manufacturers lock these inputs by default, third-party software offers a workaround. I recently purchased an application called TVQuickActions Pro to see if it could successfully reclaim these unused buttons.

Check it out in my latest video!

TvQuickActions Pro costs approximately five dollars, which positions it as a more affordable alternative to similar software like ButtonMapper.

Getting the software to intercept a button press before the Android operating system responds requires adjusting system permissions. Specifically, you must enable accessibility permissions for its core button-remapping functionality. Once configured, the software allows for granular control over individual buttons.

Instead of a standard single action, you can assign distinct commands to a single press, a double press, and a long press. During my testing, I successfully remapped Onn’s “Free TV” button to launch Plex with a single press and the RetroArch emulator with a double tap.

The application also supports visual overlays. By mapping a button to the action menu feature, you can generate an on-screen directional pad. This allows you to assign different applications to the up, down, left, and right inputs. Additionally, a dock feature can summon a customizable, horizontal menu of applications that can be navigated without returning to the primary Android home screen.

For users willing to modify deeper system settings, the tvQuickActions Pro integrates with Android’s developer (ADB) tools. By enabling Developer Options and USB Debugging in the device settings—and granting the corresponding ADB permissions within the app—you can execute system-level commands from the remote. In one example I configured a long press to force-stop the foreground application, providing a physical override to clear a suspended or problematic app from memory without digging through menus. Another developer-level feature transforms the remote into a virtual mouse, placing a cursor on the screen that can be navigated using the directional pad for applications that require point-and-click input.

The software also supports webhook transmissions, meaning a single button press could be configured to trigger external network commands, such as toggling networked room lights.

Reclaiming these locked remote control buttons provides a functional method for bypassing pre-programmed manufacturer constraints, giving you the flexibility to operate your home media setup exactly how you prefer!

Valve’s New 5 Minute One-Click SteamOS Install !

Valve’s newest SteamOS release, 3.8, no longer locks the gaming operating system to specific hardware. This means SteamOS can be installed on any PC – provided it’s running with AMD graphics hardware.

In my latest video, I install SteamOS on a Minisforum AI X1 Pro 370 Mini PC (compensated affiliate link) as a test case to see how close to an official “Steam Machine” users can get with their own hardware.

Check it out here!

Getting started was straightforward. Valve provides a SteamOS installation and repair page with the files and instructions needed to prepare a bootable external drive. On this particular PC, I had to disable Secure Boot in the BIOS before the installer would launch, but after that the process moved along without much friction.

One important limitation became clear immediately. SteamOS currently wipes the primary boot drive during installation, and it does not appear to offer a simple graphical option for choosing a different target disk. In this case, even with another drive installed, the installer defaulted to the main Windows drive. Anyone trying this should assume the main system disk will be erased unless they have taken steps to avoid that outcome.

The operating system installed in well under five minutes, and after a short post-install update and some basic setup steps for networking, display scaling, and audio output, I was at the login screen. The overall setup felt closer to bringing up a console than configuring a conventional desktop operating system.

After logging in, the interface looked very familiar. SteamOS on this PC presents the same lean-back environment that Steam Deck users already know. I moved into the display settings to confirm that a 4K panel was being detected correctly, and the system allowed me to force 3840 by 2160 output along with appropriate screen scaling without trouble. That does not mean this class of hardware is suited to modern games at 4K, but 4k is an option for lighter or older titles.

I started by downloading a game over Ethernet, where the system appeared to make full use of the PC’s 2.5 gigabit connection. From there, I loaded No Man’s Sky and began testing performance. At 1080p with the lowest settings, the game ran at a solid 60 frames per second, which is roughly where I would expect this hardware to land. Raising the resolution to 4K dropped performance to around 20 frames per second, which confirmed that 4K gaming is not realistic here for a title like this. Returning to 1080p and increasing visual quality from standard to enhanced brought performance into the 50 to 55 frame per second range, still in line with expectations.

What stood out was that the drivers appeared to load correctly, the game rendered properly, and the hardware behaved about as it does under Windows. In Cyberpunk 2077, running the built-in benchmark at 1080p on low settings, performance again matched what I usually see from this processor, with some moments looking marginally faster. There were no obvious driver issues or instability during testing.

I also wanted to see whether SteamOS on a mini PC could go beyond native PC gaming and serve as a broader living-room system. For that, I switched into desktop mode and installed EmuDeck, the software package that automates emulator setup on Steam Deck and similar systems. The process on SteamOS looked identical to the one on Valve’s handheld. After downloading the installer, dragging it to the desktop, and stepping through the setup, I had a full emulation environment in place with little manual configuration.

To test that side of the system, I loaded Burnout Revenge through Emulation Station using a PlayStation 2 emulator configured for 1080p. The result was stable 60 frames per second performance with no visible slowdowns. Controller setup was handled automatically, which is part of the appeal of EmuDeck in the first place. Users can also push emulated games directly into the Steam interface, though I tend to prefer keeping them grouped inside Emulation Station.

The final part of the test involved an external GPU. I connected a GMK eGPU equipped with an RX 7600M XT through the mini PC’s Oculink port. SteamOS recognized it immediately, and the system booted without any special configuration. I was not able to get the same unit working over Thunderbolt in this setup, although there are reports from other users of success with USB4 and Thunderbolt-connected GPUs under SteamOS.

With the Oculink-connected eGPU in place, No Man’s Sky at 4K with enhanced settings ran at about 60 frames per second. That was a meaningful step up from the integrated graphics result at 1080p, and it delivered similar to what this GPU achieves in Windows. In practical terms, the combination turned this mini PC into a living-room gaming system that boots directly into Steam’s controller-friendly interface while also handling emulation and external graphics expansion.

What emerged from this test was not a perfect universal installer, but a clear sign of where Valve seems to be heading. On compatible AMD hardware, SteamOS is already simple enough that setting up a console-style PC with an OS optimized for gaming no longer feels experimental. The main caveat is still the installer’s handling of storage, which remains blunt and potentially destructive if a user is not careful.

For anyone with spare AMD GPU hardware, or a Ryzen-based mini PC gathering dust, SteamOS now looks like a practical weekend project rather than a niche experiment. It still has clear boundaries, especially around Nvidia support and drive selection, but in day-to-day use it already feels less like a workaround and more like the early shape of a broader PC gaming platform.

Plex Hardware Transcoding on AMD Ryzen ! Zen 2 through 5 Tested

It has been four years since I last examined the feasibility of using AMD hardware for video transcoding on a Plex server. At that time, I found that hardware transcoding could be made to work on Windows with certain older mini PCs. In this month’s sponsored Plex video, I am revisiting this topic to see how the landscape has changed, specifically focusing on how different generations of AMD architecture—Zen 2 through Zen 5—perform under a Linux environment.

Check out the results in my latest video!

While Intel remains the standard recommendation for Plex hardware transcoding due to its consistent driver support and Quicksync technology, the current pricing and availability of AMD-based mini PCs make them a common choice for home lab enthusiasts. For this evaluation, I used a portable version of Unraid to test four different devices: a Zen 2-based Beelink SER4, a Zen 3 Geekom A5 Pro, a Zen 4 Geekom A8, and a Zen 5 Minix PC.

Starting with the older Zen 2 and Zen 3 architectures, the results were surprisingly functional. On the Ryzen 4800U and the 7530U, I successfully initiated hardware transcoding. When converting a 4K Blu-ray file to a 1080p 8-megabit stream, the systems utilized hardware decoding and encoding, which kept CPU utilization to a minimum. However, a significant limitation persists: hardware tone mapping is not supported on these chips. This means that while the video plays smoothly, HDR content does not display colors accurately, often appearing washed out.

As I moved to the newer Zen 4 and Zen 5 processors, I discovered hardware transcoding does not work. Instead, the systems relied on raw CPU power to handle the video processing. On the Zen 4-based Geekom A8, the 8745HS processor was capable enough to manage a 4K transcode through software, but it came at the cost of high CPU usage and noticeable fan noise. The Zen 5 Ryzen AI 365 showed similar behavior; it handled the task through software with better color accuracy for HDR content than the older chips, but lacked the efficiency of dedicated hardware acceleration.

To provide a baseline, I compared these results against a budget-friendly Intel N100 mini PC. Even as a low-end processor, the Intel chip handled hardware transcoding and HDR tone mapping simultaneously without strain. The CPU remained nearly idle while the video colors were mapped correctly, illustrating why Intel continues to be the preferred path for this specific use case.

For those who already own an older Zen 2 or Zen 3 AMD system, these devices can serve as capable Plex servers, provided the user does not require HDR tone mapping. For users with newer Zen 4 or Zen 5 hardware, the processing power is sufficient to handle one or two transcodes via software, though it is not the most efficient method.

I will continue to monitor updates to the Plex Media Server software, as future releases may eventually bridge the compatibility gap for newer AMD graphics architectures.

Disclosure: This was a sponsored post, however Plex did not review or approve this content prior to uploading.

I shoved 2 GPUs in an Old PC to Make a Killer Local AI Rig!

I spent the weekend experimenting with a six-year-old gaming PC to see if a dual-GPU configuration could transform it into a powerful local AI workstation. The objective was to determine if older hardware could handle large language models without relying on cloud-based processing. And it worked!

Check it out in my latest Local AI video.

The system is built around an Asus Maximus XI motherboard and an Intel i9-9900K processor, both dating back to 2019. It contains 32GB of DDR4 RAM.

For the graphics setup, I used an Nvidia 3080 with 10GB of VRAM (circa 2020) and added the affordable 5060ti we reviewed recently with 16GB of VRAM. Combined, these cards provide 26GB of available video memory. One notable aspect of this configuration is that it does not require specialized cabling or SLI configurations. Software such as LM Studio can detect both cards and distribute the model across the available memory just over the system bus. To maximize efficiency, I routed the display output through the Intel processor’s onboard video, ensuring that none of the GPU memory was diverted to drive the display.

During testing, I evaluated several models, including Qwen’s 35-billion parameter Mixture of Experts (MoE) model and a Google Gemma 31-billion parameter dense model. The MoE model proved to be significantly faster, generating text at approximately 52 tokens per second. Because an MoE model only activates a fraction of its parameters for any given task, it offers a higher speed of output but still needs enough VRAM to house its entire parameter space. In contrast, the dense model processed information through all its parameters simultaneously, which resulted in a slower rate of about 9 tokens per second.

I applied these models to practical tasks, such as generating code for a Space Invaders-style game and summarizing long video transcripts. The results varied by model; for instance, one model produced a visually cohesive game that required minor debugging, while another produced a stable but less detailed version. In both cases, the local setup handled the coding and iteration process in a manner similar to cloud-based AI models. I also integrated the local AI with N8N, a self-hosted automation platform, to summarize RSS news feeds. The system successfully filtered large amounts of data and generated HTML summaries with functional links in under a minute.

You can see some output samples of each these tasks here.

Finding the optimal context length was a primary technical challenge. Context refers to the amount of information the AI can retain during a conversation. While I initially used lower defaults, I eventually identified 28,000 tokens as the functional limit for this hardware. Pushing beyond this number caused the system to crash or significantly slowed performance as data spilled over into the system RAM. To maintain stability, I enabled settings to limit offloading to dedicated GPU memory and used a compression technique for the cache.

The project confirms that a multi-GPU approach can extend the utility of aging hardware for modern AI tasks. Rather than investing in a new, high-cost workstation, I was able to pool the resources of existing cards to run complex models locally. My next steps involve exploring how these models can be further integrated into daily workflows and testing their limits with more specialized data sets.

See more of my Local AI series here!

GMKTec G11 Mini PC Review – A Budget (by 2026 standards) Offering

The landscape for budget mini PCs has shifted recently. Not long ago, a well-equipped unit could be found for under $200, but those options have become increasingly scarce.

My latest Mini PC review takes a look at the GMKTec G11, a machine that attempts to balance affordability with a modern feature set. Powered by an AMD Ryzen Embedded R2514 processor, the unit I tested includes 16GB of RAM and Windows 11 Pro for a fairly reasonable price. For those who prefer to provide their own components, a barebones version is available for about $169 (compensated affiliate links).

The hardware design is functional, offering a variety of connectivity options. On the front, there are three 10Gbps USB 3 ports, including one Type-C port that supports 4K video output at 60Hz. It is worth noting that this port does not support power delivery, so the included external power supply is still required. The back of the device features dual 2.5 gigabit Ethernet ports, which performed to standard in my testing, and two additional HDMI ports that can each deliver 4k 60 independently. For wireless connectivity, the G11 uses a MediaTek Wi-Fi 6E radio. In my environment, it reached speeds of 600Mbps downstream and nearly a gigabit upstream.

Inside the chassis, the G11 shows some improvements over previous entry-level models from this manufacturer. It now features two NVMe slots rather than one running with the older mSATA standard, allowing for dual-drive configurations. The RAM is also user-upgradeable, with two DDR4 slots supporting up to 32GB.

Cooling is handled by an internal fan that remains relatively quiet during standard operations. Users can toggle between three performance modes in the BIOS; switching from the default 20-watt “standard” mode to the 25-watt “performance” mode provides a measurable boost in benchmarks, though it does cause the fan to engage more frequently.

My initial experience with the software was less straightforward. Unlike previous GMKTec machines I have reviewed, which arrived with a clean installation of Windows, this unit included several pre-installed applications. One, called “GMKTec Claw,” appears to be a version of an AI application that requires a login I did not have. Another, titled Herdsman, is apparently a Local AI client that was entirely in Chinese. Given the low-level system access these types of AI tools often require, I would suggest that users consider a fresh installation of Windows or a move to a Linux distribution upon receiving the device.

In terms of daily performance, the G11 handles standard productivity tasks, such as web browsing and office applications, without difficulty. However, I encountered some limitations during specialized testing. While streaming 4K video at 60fps on YouTube, the system dropped a noticeable number of frames, suggesting that the drivers for this specific Ryzen processor may still require optimization. The Browserbench.org Speedometer benchmark placed the R2514 in the same performance bracket as Intel’s N300 and N150 processors, which are common in this price range.

Graphical performance is modest but better than its low-end Intel competitors. During a test of No Man’s Sky at 720p on the lowest settings, frame rates stayed consistently below 30fps. Similarly, PS2 emulation struggled to reach full speed at native resolutions.

For those interested in media serving, the AMD architecture presents some challenges with Plex. While I managed to get some hardware transcoding working in Windows, higher-bitrate 4K HDR files caused the CPU usage to spike and the video to buffer. This is an area where an Intel Mini PC generally maintains an advantage.

But the G11 proved to be a stable platform for Linux. I booted the latest version of Ubuntu and found that the hardware was recognized immediately, including the Wi-Fi and Bluetooth modules. The operating system felt responsive, and for users looking to run Docker containers or a basic home server, this hardware offers a capable foundation. The system draws about 11 to 12 watts at idle and peaks at 48 watts under heavy load, keeping temperatures around 60°C while running sustained heavy workloads.

The G11 represents the current entry point for the mini PC market. While it no longer carries the sub-$200 price tag of years past for a full configuration, it provides a functional set of ports and upgradeable internals that suit basic desktop needs or lightweight server duties.

Amazon Tech Haul 16! Prime Day 2026 Edition

It’s time for another gadget tech haul! This time I have a collection of seven new items that I found of interest ahead of Amazon’s 2026 Prime Day(s)! Check out the video here!

You can find a compensated affiliate link to the complete list here. Many of the items are on sale for Amazon’s big summer promotion. With the exception of the Logitech Mouse below, all of these items came in free of charge from their manufacturers. All links below are compensated affiliate links.

The first device is the Aurzen D1R Air portable projector. It operates on the Roku TV platform and weighs approximately 2.6 pounds. The unit projects a 300-lumen image, which is optimized for an 80-inch display in darkened environments. It supports standard streaming applications and wireless display casting from computers, though the integrated speaker is basic and the input lag makes it less suitable for fast-paced video games. The device features automatic keystoning and focusing, and it can be powered via a traditional wall outlet with its included power cable or a 65-watt USB-C power supply (not included). The manufacturer also provided me with an optional tripod accessory equipped with an internal battery capable of powering the projector for roughly two and a half hours.

The Logitech Mobi Fold is a full sized bluetooth mouse that is optimized for travel thanks to its foldable design. It features standard left and right buttons, a capacitive scroll area, and forward/back navigation controls built into the center scroll surface. It charges via USB-C, pairs with up to three separate devices, and includes a user-replaceable battery.

Anker’s Soundcore Work is a meeting recorder that has server-based AI features for transcription and summarization. The hardware features a detachable, wireless microphone that magnetically clips to clothing, with the base unit able to attach to phones via Magsafe. Users receive 300 free minutes of transcription per month, after which a subscription fee applies. The device can be used without a subscription, but audio extraction is currently limited to the manufacturer’s proprietary app only.

In the diagnostic category, Treedix sent along their awesome USB Cable Tester. This tool identifies the specifications of connected cables, such as USB 4 or Lightning, by reading pinouts and e-marker data. It can also conduct a resistance test. It operates on a single AAA battery or direct USB-C power.

The Zima Board 2 is a fanless mini PC/server equipped with an Intel N150 processor, 8 gigabytes of non-upgradeable RAM, and 32 gigabytes of eMMC storage. It features a PCIe slot on the side for adding expansion cards (even GPUs), along with two powered SATA connectors for attaching external storage. It utilizes a custom operating system designed to simplify Docker container installations and does so more effectively that many competing NAS devices. However, the software defaults to older versions of these applications, requiring manual intervention to secure the latest updates. It’s priced higher than other comparably spec’ed Mini PCs that have more upgrade paths. But I found its headless design and Zima OS to be interesting enough to include in this lineup as there is potential here.

For portable power, Sharge sent over their Shargeek 170 power bank. It features a transparent casing that exposes its circuitry and battery cells, a 24,000 milliamp-hour capacity, and can deliver a maximum output of 170 watts. The built-in display provides real-time data on wattage and remaining runtime. In testing, it successfully powered the Aurzen projector and charged a dead mobile phone simultaneously while accepting an incoming charge. Plus it looks super cool.

The last items on the list is the ClonerAlliance Box Pro Pocket, a standalone video capture device. It records up to 1080p video at 60 frames per second via HDMI, but its standout feature lies with its analog inputs for RCA, component, and VGA connections. In the video I demonstrated capturing the VGA output from my 29 year old college laptop along with a VHS capture and upscale to 1080p. The device features an integrated screen and saves video directly to external storage (USB or SD card), providing a straightforward method for digitizing legacy media formats.

If you liked this lineup I have over a dozen more similar hauls I’ve done over the last couple of years! Check them out here.

Lenovo Idea Tab Plus 2026 Review – Budget Android Tablet

With technology prices rising due to memory shortages, finding a capable tablet that remains affordable has become more of a challenge. I recently reviewed the Lenovo Idea Tab Plus, a 12.1-inch device that is currently bundled with a case and pen at a great price over at Amazon (compensated affiliate link).

Click here to see my video review!

The tablet features a metal back and a glass display, weighing in at 1.17 pounds. While it supports accessories like a stylus and a keyboard case via pogo pins, these are not included in the base configuration which is what Lenovo sent to me for review.

The 12.1-inch screen offers a resolution of 2560 by 1600 and reaches a brightness of 600 nits. With a 90Hz refresh rate and 96% DCI-P3 color coverage, the interface feels responsive and the color accuracy is sufficient for general media consumption and light photo editing.

Under the hood, the device runs on a MediaTek Dimensity 6400 eight-core processor paired with 8GB of RAM. For storage, it provides 128GB internally, though an SD card slot allows for expansion up to 2TB. The physical ports are limited to a single USB-C 2.0 port used for charging and data; notably, it does not support wired video output.

During my daily use, the battery life consistently lasted between 10 and 12 hours for standard tasks like web browsing and email. The audio experience is supported by a quad-speaker system that automatically adjusts its stereo orientation when switching between landscape and portrait modes. While the 13-megapixel rear camera lacks stabilization, it produces sharper still images than many other budget tablets I have tested. The 8-megapixel front camera is adequate for video conferencing and supports facial recognition for unlocking the device, which is necessary as there is no fingerprint reader.

Performance is steady for multitasking. I found that I could run a YouTube video alongside a web browser without significant lag. The tablet also supports HD playback on streaming services like Netflix, which utilizes the 2K display effectively. For gaming, the hardware is capable of handling standard Android titles and some Gamecube emulation through the Dolphin emulator, though it does not match the benchmarks of more expensive models like the entry-level iPad or even some recent high-end budget competitor from Walmart’s Onn brand.

Two software features stood out during my time with the device. The first is a “PC mode” that allows apps to run in resizable windows, mimicking a desktop environment. This is particularly useful if the tablet is paired with a Bluetooth mouse and keyboard. The second is an integrated GPS radio that functions without an internet connection, a feature increasingly absent from other tablets in this price range.

On the software side, the device ships with Android 16. Lenovo has committed to two operating system updates and four years of security patches. However, the initial setup experience is somewhat hindered by a significant amount of pre-installed bloatware and prompts to download additional sponsored applications. Once these are cleared and the initial system updates are finished, the performance levels out.

While it lacks the high-end power found in flagship devices, this tablet occupies a practical middle ground. It offers a solid build and a high-resolution display for users who prioritize media and productivity over heavy gaming.

Disclosure: Lenovo provided the tablet for review on loan. No other compensation was received and they have not reviewed or approved this content prior to publication.

See more tablet reviews here!

This Retail GPU Works Great for Local AI !

I recently began exploring the practical applications of local AI by integrating a new GPU into my hardware configuration. I purchased an Asus card equipped with 16 gigabytes of video memory (compensated affiliate link) that by today’s standards is at a fairly “budget” price.

Check out what it can do in my latest video!

16 gigabytes is enough RAM to house the Gemma 4 26-billion parameter mixture of experts model. This specific model provides a balance of performance and efficiency that mirrors what many users expect from cloud-based subscriptions, but without the need for external data processing and costs.

My current setup involves an external GPU configuration using an Oculink Thunderbolt dock connected to a GMKTec mini PC. While the computer itself contains 64 gigabytes of system RAM, running these models exclusively on a GPU is necessary for maintaining acceptable speeds. For the software interface, I am using LM Studio, a free cross-platform application that allows for fine-tuning performance settings. One useful feature of this software is the ability to offload portions of a model to system memory if the GPU’s capacity is exceeded, though this results in a significant reduction in processing speed.

During initial testing, the Gemma model generated text at a rate of approximately 45 tokens per second generating a short fiction story. While the system consumes about 200 watts of power under a full load, it idles at 36 watts. Beyond simple text generation, the setup is capable of visual analysis. In one test, I provided the model with a photograph of some friends and I in front of Space Shuttle Atlantis. It accurately identified the shuttle and the attributes of the individuals in the frame, correctly processing the visual data without external assistance.

I also tested the model’s ability to handle complex document analysis. I combined a 24-page FCC proposal regarding prepaid smartphones with a transcript of a video I had previously recorded on the topic. Because LM Studio does not currently support PDF files, I converted the information into a single text document. After increasing the context length to its maximum setting to ensure the AI could “see” the entire file, I asked it to draft testimony for the FCC based on my specific concerns raised in the video and attributing those concerns to a specific portion of the FCC’s draft proposal. The model successfully identified my points regarding the privacy of whistleblowers and reporters and was mostly correct in its attributions.

The model demonstrated further utility in coding tasks. It managed to write a functional browser-based Space Invaders clone in a single attempt, including basic game logic and sound effects. Later, I used it to generate a Python script designed to scrape education statutes from a state website. When the first version of the script returned an error, the AI analyzed the problem and provided a corrected version that successfully consolidated numerous legal chapters into one searchable document.

For datasets too large for a standard chat interface, I utilized an application called Anything LLM to perform retrieval augmented generation, or RAG. This process involves indexing and embedding documents locally so the AI can query them efficiently. I uploaded five megabytes of state statutes and asked the model to calculate a specific school grant amount based on a complex formula. The model performed the necessary math and returned the correct figure of $2.2 million. I found that while the local model sometimes requires more specific prompting than cloud-based alternatives like Google’s Notebook LM, it is capable of providing high-fidelity results.

Running larger models, such as those with 31 billion parameters, reduces performance to about five tokens per second when the GPU memory is exceeded. However, the flexibility to swap between various models from Google, Qwen, or other developers allows for a customized approach to different tasks. These tools have reached a point where they are functional for data analysis and automated workflows while keeping all information on a local server. For anyone with a modern video card and a sufficient amount of VRAM, these local models offer a viable way to experiment with AI without relying on the cloud.

Fox to Acquire Roku in a $22 Billion Deal.. What’s Next?

The landscape of streaming TV shifted Monday with the announcement that Fox Broadcasting is set to acquire Roku. What will the implications of this $22 billion deal be for the average television viewer?

In my latest video, I dive into the details the two companies released during an investor call.

Roku’s current position in the market is a primary driver for this acquisition. The company currently accounts for 44% of total hours spent viewing connected TV content by brand in the United States. This dominance is not just limited to their standalone streaming sticks but extends to the numerous television sets that use the Roku operating system.

Over the last year, Roku reported approximately $5 billion in revenue, with advertising accounting for half of that figure. The business has transitioned away from hardware, which now represents only about 11% of their revenue, toward a model built on ad sales, data collection for targeted marketing, and a share of subscription fees from third-party services like Netflix.

For Fox, the acquisition provides a path to the top tier of the streaming market. By combining its existing viewership with the Roku Channel, Fox moves into a competitive position alongside industry leaders, trailing only YouTube in terms of market share and gaining ground on Netflix and Paramount. The company expects significant growth in both connected TV advertising and streaming subscriptions through this integrated platform.

Based on the details provided in the investor call, users should expect noticeable changes to the Roku interface. One of the key shifts involves the promotion of Fox-owned content and properties. The current “top picks” and algorithmic recommendations will likely favor Fox programming over other partners. The sports section is also expected to feature a heavier concentration of Fox Sports content. There is a possibility that the traditional grid of apps, which many users prefer for its simplicity, may be moved further down the screen to make room for these promoted recommendations and additional advertising spaces.

While Fox representatives stated they intend to maintain a level of “neutrality” toward other providers like Disney—partly because Roku earns revenue from selling those subscriptions—the emphasis on organic-looking Fox recommendations suggests a more curated experience. The core functionality of the devices is expected to remain intact, as the value of Roku has always depended on its ability to host a wide variety of streaming services. However, the path a user takes to find their preferred apps is likely to require navigating around Fox’s priorities.

The acquisition is structured as a mix of 60% cash and 40% stock, valuing Roku at $160 per share. Once the deal is finalized, Fox shareholders will own 73% of the combined entity, while Roku shareholders will hold the remaining 27%. Anthony Wood, the founder of Roku who previously established ReplayTV, is slated to remain on the board.

The deal still requires approval from shareholders and government regulators. with the companies anticipating the deal closing sometime in 2027. Watching how the interface evolves as we approach the closing date will provide the best indication of what the future of this hardware will look like in a Fox-owned ecosystem.

Let’s Talk About Windows 11’s Bloated Notepad and a Fix!

A few weeks ago, I published a video about Microsoft’s recognition about how bloated Windows has become. You don’t have to look much further than the current status of the Notepad app to see that bloat in action.

In my latest video, we take a look at Notepad.exe’s feature creep and how you can download a free alternative that brings the simple version we have all come to know and love back.

Upon opening Notepad now, the software automatically loads every document previously accessed, syncing them across a Microsoft account and putting each one in its own tab. It includes formatting options, a tabbed interface, and even integrated AI tools like Copilot.

This shift moves away from the utility’s original purpose. When I first used Notepad on Windows 3.1 in the early 1990s, it was a straightforward tool for taking quick notes. It handled one document at a time, requiring the user to close one file before opening another. This simplicity remained largely unchanged through several iterations of the operating system, from the original Windows 1.0 through Windows 10. The utility functioned effectively for decades because it remained focused on basic text entry without additional overhead.

The technical demands of the current version are also notable. In my recent observations, running four small text files in the updated Notepad consumed roughly 87 megabytes of RAM. This footprint appeared to grow the longer the application remained open, adding 10 megabytes during my testing period without opening any additional files, which may suggest background telemetry or memory management issues. To put this in perspective, the entire computer system I used in 1992 operated on only 4 megabytes of total memory. While some users might find the added features beneficial, the increased resource consumption seems high for a simple text editor – especially as Microsoft’s more robust text editor, WordPad, has been removed with its features now added to Notepad.

For those seeking a return to a more minimalist experience, an alternative has emerged from the open-source community. Dave Plummer, a retired Microsoft software engineer, has developed an application called RetroPad. It is designed to function like the classic version of the software, prioritizing speed and a minimal footprint. In a direct comparison, RetroPad used approximately 1.7 megabytes of memory to handle the same text that required nearly 88 megabytes in the official Microsoft version.

The application is currently available via GitHub, where the executable can be downloaded from the project’s binaries folder. Plummer is also developing a version in assembly language to further reduce the executable size, but I noticed its memory footprint was significantly larger than the other version.

For users who find the modern updates to be an unnecessary complication, this project offers a functional way to return to a simpler workflow.

Is the FCC Coming for Burner Phones & Prepaid Accounts?

I have been following reports regarding the Federal Communications Commission’s stance on the future of prepaid mobile devices and sim cards, often referred to as “burner phones.” While some headlines suggest a total ban is imminent, the reality is the FCC is currently weighing requiring that individuals identify themselves with a government ID when purchasing these devices and plans.

In my latest video, I take a look at what the FCC is considering and what the implications might be.

The core of this issue lies in the persistence of illegal call traffic. Many of us are familiar with the daily influx of fraudulent text messages and automated calls, ranging from “pig butchering” investment schemes to fake government notifications. To facilitate these operations, bad actors frequently utilize “SIM farms”—setups where hundreds of SIM cards are loaded into machines to send out bulk messages efficiently. Currently, it is relatively easy to walk into a convenience store, pay cash for a prepaid SIM card, and gain access to the cellular network with total anonymity.

Under a new proposal, the FCC is looking to tighten the requirements for these purchases. The proposed rules would require providers to obtain a customer’s name, physical address, and a government-issued ID number, along with an alternate phone number, before activating service. This would apply to both new and renewing prepaid customers. Furthermore, the commission is considering disqualifying the use of P.O. boxes, mail forwarding services, and shared office locations as valid addresses, arguing that these do not sufficiently verify a user’s identity.

There are, however, significant privacy implications to consider. Certain groups rely on the ability to remain anonymous for their safety and professional integrity. This includes whistleblowers, domestic violence survivors who need to keep their location hidden from abusers, and journalists working on sensitive assignments. By requiring a government ID and a physical home address for every prepaid phone, the barrier to maintaining that privacy becomes much higher.

Groups like the Electronic Frontier Foundation have expressed skepticism about the effectiveness of such a move. The argument is that criminals will likely not be deterred by these requirements; instead, they may simply use stolen identities from the numerous databases already available on the dark web to bypass the system. This creates a risk where the policy could inadvertently incentivize more identity theft while doing little to stop the actual robocalls.

I find the focus on the end-user somewhat misplaced. Mobile carriers have the technical capacity to identify when hundreds of SIM cards are operating from a single location and connecting to the same cell tower to blast out thousands of messages. Concentrating on carrier accountability and the detection of SIM farms might address the root of the problem more effectively than collecting the personal data of every individual user.

Within the commission, there is a visible divide on how to proceed. While the two Republican commissioners (Chairman Brendan Carr and Commissioner Olivia Trusty) have labeled this a top priority for consumer protection, Democrat Anna Gomez has raised concerns regarding the privacy of those who use prepaid services legitimately. Gomez feels the onus should be on the carriers to detect fraudulent traffic.

As this proposal moves through the rule-making process, the commission is seeking public comment on dockets 17-59 and 02-278. In the video I demonstrate how you can submit public comment on this docket.

So while this is not the end of burner phone, the era of anonymous entry into telecommunications network may be nearing an end. I will continue to monitor the docket as this proceeding makes it through the process.

See more analysis pieces like this here!

Local AI is Finally Usable! Real-World Workflows on a 5-Year-Old Mac with Google Gemma

For the past few years, I have been downloading local AI models to determine if they could handle practical automation tasks or summarize long-form content. Historically, these experiments have been unsuccessful, with the models typically failing to provide comprehensive results or losing track of the original context. However, recent developments in model optimization, specifically on Apple hardware, have changed the baseline for what is possible on a personal computer.

In my latest video, I demo running Google’s latest Gemma 4 model, a 26-billion parameter “mixture of experts” model optimized for the Mac using the MLX framework on my 2021 Macbook Pro with an M1 Max and 32GB of RAM.

Check it out here!

I observed the model generating approximately 50 tokens per second. While this is slower than a high-end cloud-based system, it represents a very usuable generation speed for a local setup. The unified memory architecture of the Mac allows the GPU to access data efficiently, which is why these older machines remain relevant for AI tasks that would otherwise require significant cloud computing resources.

During my testing, I provided the model with a transcript from a recent video to see if it could produce a coherent summary. Unlike previous local models that often provided incomplete or erratic responses, this model maintained a consistent narrative and adhered strictly to the provided text. I also tested it with a dense legal document from an FCC docket. After processing a large amount of extracted text, the model was able to delineate the key arguments of the filing and, upon further prompting, condensed the information into a concise executive summary.

I also examined the model’s vision capabilities using a tool called MLX studio, which supports image analysis. I uploaded a photograph with some friends and I in front of a space shuttle and asked the model to describe the scene. While it misidentified the vehicle as a Dreamchaser—a different type of spacecraft—the level of detail was a step forward from earlier local models that often provided much less accurate descriptions. This functionality is particularly useful for my ongoing project to index a large archive of digital photos dating back to 1997. Using a local model for this type of organization could potentially eliminate the costs and privacy issues associated with the thousands of API calls required for cloud-based indexing.

To test the model’s utility in a production environment, I integrated it into my N8N automation server. I currently use a cloud-based AI with my N8N server to scan news feeds and identify relevant stories for my daily work. I ran a portion of this same workflow using the local Gemma model to see if it could replicate the results. It took approximately three minutes to process the news briefing. Although the results were not quite as polished as those from the cloud, the model successfully identified unique stories and avoided duplicated stories about Apple’s WWDC event that were being published at the time.

Google appears to be prioritizing the development of effective local models more than some of its competitors, providing a way for users to utilize AI without incurring expenses beyond the electricity required to run their own hardware. Seeing a 26-billion parameter model function with this level of stability on a five-year-old laptop has caused me to rethink my existing workflows. I am now looking at which of my daily tasks can be moved away from the cloud and managed entirely on my own hardware.

See more videos like this here!

ATSC 3.0 Update: DRM Breaks TV Emergency Messaging!

Last week, Weigel Broadcasting met with the Federal Communications Commission (FCC) to present the findings of a study they conducted regarding encrypted TV signals and Emergency Alert System (EAS) messages . They found that the EAS messages were encrypted along with the content, limiting who can actually see it.

In my latest video I detail Weigel’s findings.

For the past two decades, the ATSC 1.0 standard has allowed nearly any television to receive TV broadcasts with a simple antenna. These devices do not require firmware updates or internet connectivity to decode signals; they operate on a settled standard that works immediately upon being plugged in. In contrast, many ATSC 3.0 televisions and tuning boxes require digital certificates to decrypt content. In some cases, these certificates require an internet connection to refresh, creating a potential point of failure during a crisis – especially if the user doesn’t regularly use the over the air tuner on their television.

Weigel’s presentation to the FCC highlighted a documented disconnect between industry claims and technical reality. While some industry groups have previously stated that content protection does not impede the delivery of emergency alerts, Weigel’s testing in Chicago, Charlotte, Green Bay, and South Bend suggested a different outcome. Using both certified and uncertified tuning devices, researchers found that if a device was unable to decrypt the primary television signal, it failed to display the Emergency Alert System (EAS) messages. The emergency information was essentially trapped inside the encrypted signal.

This creates a significant reliability issue. During severe weather events or other catastrophes, internet and cellular infrastructure are often the first services to fail due to damaged fiber optic lines or power outages. If a television requires a handshake with a remote server to decode a local broadcast, the very system designed for emergency redundancy becomes dependent on the infrastructure most likely to be compromised. I observed a practical example of this earlier this year when a certified tuning box failed to decode encrypted channels during a blizzard due to a security certificate error, even though it was marketed as not requiring an internet connection.

The financial and technical barriers to entry are also notable. Currently, ATSC 3.0 tuners are primarily found in high-end television models, leaving low- and mid-range consumers without native access. The third-party boxes that can bridge this gap are subject to a certification process controlled by the nation’s largest broadcasters. This allows those broadcasters to determine which devices are permitted to decrypt the public airwaves, a shift from the open-standard approach that’s been around for decades.

Weigel’s recommendations to the FCC seek to establish several safeguards as this transition continues. They are advocating for the continued requirement of ATSC 1.0 simulcasting for the foreseeable future to ensure the public is not abruptly cut off. Furthermore, they suggested that a minimum of 19.3 megabits per second of bandwidth be dedicated specifically to free over-the-air video to prevent spectrum from being diverted to private data services. Their proposal also concedes that while digital rights management may exist, it should never require an internet connection for decryption, and signal reception thresholds should remain consistent so consumers are not forced to replace existing antennas.

The FCC is now faced with a choice regarding how to balance the revenue desires of broadcasters with the safety requirements of the general public. The data provided by this study offers a specific look at the risks of prioritizing revenue over accessibility. What’s next? Stay tuned!

See more in my ATSC 3.0 series here.

SuperStation One With Dock MiSTer Review : Lots of Analog TV Options!

After months of waiting, I finally got my Superstation One, a new “consoleized” MiSTer device developed by hardware maker Taki Udon. For those who follow the MiSTer open-source project, the Superstation One represents a shift toward a more consumer-friendly form factor that does not require stacking a bunch of circuit boards together.

See it in action in my latest MiSTer review!

For those not familiar, a MiSTer runs classic computers and video games (both console and arcade) using FPGA technology. The FPGA chip replicates the logic of the original chips found on these devices which makes it incredibly accurate to the original. While the Superstation One looks a lot like the original Playstation, complete with controller and memory card slots, it is compatible with the full MiSTer library of old computers and gaming platforms.

The Playstation ports utilize an integrated “SNAC” interface to connect the peripherals directly to the PS1’s FPGA core, which reduces input latency compared to standard USB. While the base unit is limited to PlayStation controllers, adding the dock provides a universal SNAC port for other adapters, such as those for NES Zapper light guns as I demoed in the video.

The base console retails for $210, and I paired mine with the $45 docking station. The dock includes an optical drive intended to eventually boot physical media for systems like the Sega CD and TurboGrafx-16. However, at this stage of development, the CD drive is not functional within the software. The dock does provide other useful features, such as an NVME slot for high-capacity SSD storage, several additional USB ports, USB interfaces for development and a “SNAC” port for attaching original controllers and light guns.

The variety of video outputs sets this apart from other MiSTer clones I have used. It includes a standard HDMI output found on most modern devices, but it also integrates a wide range of analog video options: VGA (including the option to break VGA out to component RGB), a Sega Saturn-style DIN connector (for special DIN RGB cables), a dedicated component video output, S-video, and composite RCA. This is currently the only MiSTer that incorporates all of these video options directly on the hardware.

In my testing, I was able to run an RCA composite signal to a 40-year-old Hitachi television while simultaneously outputting an HDMI signal to a modern monitor. This allows for playing on a period-accurate zero latency CRT display while capturing high-definition footage for a stream or recording. The video signals were clean, and the composite and S-video options worked without requiring additional configuration (for most cores) or external dongles.

I tested several demanding cores, including the PlayStation, Saturn, Neo Geo, and Sega 32X. Everything performed the same as other MiSTer hardware I have tested.

The software environment, however, feels unpolished. Because this hardware is offering video output features beyond what is typically found on a stock MiSTer, I encountered some cores that required specific dip-switch adjustments to output color correctly over certain analog connections. Taki Udon is developing a custom, console-style menu system that is currently in a beta state that is just barely usable. Eventually that interface will be able to access the optical drive but for now it’s mostly a proof of concept.

The build quality is generally solid, though there are minor points of concern. The plastic doors on the unit can detach easily, and the system runs noticeably warm during operation. While my unit remained stable during extended memory tests without a cooling fan, the developer now offers an active cooling solution for those concerned about heat.

But all in, this hardware offers a more accessible entry point for users who want a turnkey solution rather than a DIY assembly. It consolidates features that usually require multiple add-on boards into a single footprint. As Taki now focuses in on the software aspects fo the device and the optical drive becomes functional, the utility of the Superstation One will expand.

California Backing Off on Linux Age Verification ?

I recently followed up on the developments regarding California’s age verification law, which originally sought to require all operating system providers—including those in the open-source community—to collect the birth dates or ages of their users. A new amendment is currently moving through the California legislature that attempts to address some of the concerns raised by the initial language, specifically by creating a carveout for open-source providers.

We take a look at that in my latest video.

The proposed amendment adds a section to exempt any person or entity that distributes an operating system or application under license terms allowing a recipient to copy, redistribute, and modify the software. Under this definition, open-source projects would likely be exempt from the data collection requirements. However, the scope of this exemption raises questions about major platforms like Android. While Google layers closed-source services onto its devices, the underlying operating system is redistributable and modifiable, which could potentially place it within this new exemption.

Despite the move to protect open-source software, other parts of the amendment introduce new privacy considerations. The updated language specifies that operating systems must share a user’s age information with several third parties, including application stores, developers, browser providers, and internet website operators. Because website operators in California may be held accountable for the age of their users, they are likely to request this data as a matter of course to ensure they are in compliance with the law.

The amendment also places the onus on developers and website operators to act on their own data should it conflict with the age information sent by the operating system. This suggests that the data provided by the operating system may not be treated as a definitive source, which could encourage websites to collect even more personal information independently to avoid legal trouble. Furthermore, the law does not currently appear to restrict these developers or website operators from sharing that age data with other third parties.

This legislative activity passed the California State Assembly last week by a vote of 68 to 1 and is now heading to the State Senate. It is expected to be signed by the governor before the original law takes effect this fall. This is part of a broader national trend, as seen in Texas where the state is currently taking legal action against platforms like Discord to enforce mandatory age verification.

As state governments continue to move forward with these requirements, the impact on individual privacy remains a primary point of discussion. While these laws aim to regulate access, they may also lead to a shift in how people use the internet. I will be watching to see if these measures result in a rise of decentralized, open-source applications that replicate the functionality of major social platforms while operating outside the reach of centralized data collection mandates.

Big Changes Coming to Android Sideloading

I have long valued the Android operating system for its openness, particularly the ability to sideload applications outside of the Google Play store. Whether I am installing a PlayStation 2 emulator on a budget tablet or adding specialized benchmarking tools to a Google TV device, the process of downloading an APK file and bypasssing the official storefront has been a hallmark of the platform.

But this flexibility is set to put under significant restrictions as Google prepares to implement a new policy regarding how apps are distributed and installed outside the Play Store.

See more in my latest analysis video!

Under the new rules, developers who wish to offer apps for sideloading will be required to verify their identity with Google, providing legal names, addresses, and official government identification. For organizations, this includes providing a Dun’s number and a verified website. Even if an app is hosted on a private website rather than the Play Store, it must be registered with Google and signed with a private key. While there is a limited distribution tier for small-scale projects involving fewer than 20 devices, broader distribution will require a registration fee, currently set at approximately $25.

For the average user, the process of installing an app that hasn’t been registered through this process is becoming notably more complex. When attempting to sideload an unsigned application, a user must first confirm they are not being coached by a third party to disable security settings. This is followed by a mandatory device reboot and full reauthentication of their Google account to terminate any potential remote access or active calls. Most notably, a 24-hour waiting period is then triggered. Only after this day-long delay can the user return, verify their identity via biometrics or a PIN, and finally complete the installation.

These protections can be enabled temporarily for seven days or indefinitely on a per-device basis. Google maintains that these steps are necessary to combat rising instances of malware and social engineering, citing examples like banking fraud and malicious software disguised as wedding invitations. By adding these layers of friction, the company aims to protect vulnerable users who might be pressured by scammers into installing dangerous software. Yet, for those who understand the risks and simply want to maintain control over their hardware, these changes introduce a substantial inconvenience.

Opposition to this shift is already organizing under the “Keep Android Open” banner. This group argues that Google is retroactively locking down an operating system that was originally marketed as the open alternative to Apple’s closed ecosystem. A point of concern for many is that these changes are not being implemented through a standard Android OS update. Instead, they are being rolled out via Google Play Services. This means the new rules can be applied to billions of existing devices without a full firmware overhaul. The rollout is scheduled to begin this month in regions including Brazil, Indonesia, Singapore, and Thailand, eventually expanding globally.

The tension here lies between individual autonomy and collective security. While Google’s stated intent is to minimize liability and protect users from financial theft, the methods chosen may fundamentally alter the relationship between the user and their device. I would prefer to see these options managed at the account level, allowing experienced users to opt out of the 24-hour waiting period while keeping protections active for others. As these policies take hold, we may see a rise in interest for alternative, open-source mobile operating systems like GrapheneOS, which prioritize privacy and side-loading without Google’s oversight.

Using AI to Organize and Curate My Plex Library ! (sponsored post)

Many media server enthusiasts like me spend more time managing our libraries than actually consuming the content within them. I recently looked into whether artificial intelligence could streamline this process, specifically within the Plex ecosystem.

In my latest monthly sponsored Plex video, I install a Plex MCP and connect Open AI’s Codex to see what we can do.

My goal was to see if AI could handle the organizational tasks that usually require manual input, such as curating specific story arcs or generating recommendations based on viewing history.

The bridge between the AI and the media server is a Model Context Protocol (MCP) server. MCPs in this context connect an AI desktop client—such as OpenAI’s Codex, Claude Desktop, or Gemini—to a local Plex instance. Setting this up requires a Plex token, which is accessible through the XML info of any media item in the library. Once the token is dropped into the configuration screen, the AI can interact with the server’s data through plain English commands. I demo this in the video.

One practical test involved creating a playlist for a specific narrative arc within Star Trek: Deep Space Nine. Rather than manually selecting episodes that contribute to the “Dominion War” storyline, I provided the AI with a text file containing an episode list sourced from a Reddit discussion. The system generated a sequential playlist in approximately ninety seconds. Beyond simple lists, the AI can also aggregate data from the web. I asked it to research the highest-rated episodes of Star Trek: The Next Generation across multiple sources and build a playlist based on those aggregate ratings. While this research-heavy task took several minutes, it resulted in a curated selection based on external critical consensus without any manual sorting or list creation on my part.

The utility extends to music as well. I requested a smart playlist for 1980s pop music. By defining the decade and genre in a simple prompt, the AI configured the metadata filters within Plex to include tracks from 1980 to 1989 categorized as pop or rock. Because it was set up as a smart playlist, it automatically updates whenever new tracks fitting those criteria are added to the library in the future. This replaces the need to navigate through various menu layers to set up complex filtering rules.

Analysis of viewing habits is another area where the AI provides utility. By accessing watch history, the system can identify content that has been overlooked. For instance, I asked it to find episodes of a specific children’s show, Bluey, that had the lowest view counts to find episodes my children might have missed. It can also cross-reference a watch list with a viewing history to suggest new series. When I requested science fiction recommendations, the AI suggested several shows I had not yet watched, such as Farscape and Babylon 5. It then added these items to my Plex watch list upon request and identified which streaming services currently host them.

For those interested in more advanced automation, the AI can be used to monitor upcoming releases of TV shows and movies. I currently use a routine that scans my watch list every morning and sends an email notification when new episodes of my tracked shows are scheduled to return to streaming services or broadcast.

By using an AI as an intermediary, the technical burden of database management is replaced by a conversational interface. It’s a definite time saver and fun to play with!

See more of my Plex content here!

Walmart Onn Pro 13.2″ Tablet Review

I recently purchased the new Onn Pro 13.2-inch Android tablet from Walmart for $288 (compensated affiliate link) to see how it compares to other devices in the budget-conscious market.

In my latest video review, I take the tablet out for a spin and check out all of its features – some of which aren’t even advertised.

The tablet runs on Android 16 and features a 13.2-inch display, which is a significant size for a device at this price point. It feels substantial, weighing approximately two pounds, with a metal design and a glass display that feels thick and durable, though it lacks an anti-fingerprint coating and tends to collect dust and smudges.

The hardware includes a MediaTek 8775V processor, 8GB of RAM, and 256GB of internal storage, which can be expanded via a micro SD card slot. It carries an IP54 rating, meaning it is resistant to dust and splashes but should not be submerged.

The display has a 2400 by 1600 resolution with a 3:2 aspect ratio. While the specifications do not list a high refresh rate, I found that the settings allow for 90Hz or 120Hz, providing a more fluid visual experience than the standard 60Hz found on most entry-level tablets. However, the included pen does not function when the screen is set to 120Hz. The brightness appears to be between 400 and 450 nits, which is more than adequate for indoor use but lower than higher-end iPads.

A notable feature is the USB-C port’s ability to function as a video input. When I connected the tablet to my Mac, it prompted me to use the device as an extended display. While this worked, the tablet currently struggles with aspect ratio mapping, resulting in a squished image that distorts the desktop. There is some minor latency, but the feature shows potential as a portable monitor if future firmware updates address the scaling issues. It is important to note that the port only supports video input, not output, so the tablet cannot be mirrored onto a larger television or monitor via cable.

The tablet comes with a basic protective case and a USI 2.0 compatible pen. The pen attaches magnetically to the top of the tablet, though it does not charge there; it requires a separate USB-C connection for power. Because it follows the USI 2.0 standard, users can swap it for other compatible third-party pens. In my testing, the writing experience felt functional for note-taking or basic sketching, though there was some noticeable lag and occasional issues with palm rejection when the pen was lifted too far from the screen.

In terms of daily performance, the device handles web browsing and multitasking efficiently. I was able to run split-screen applications, such as YouTube alongside a browser, without any significant slowdown. The software environment is relatively clean, with the Walmart app being the only pre-installed software beyond standard Google services. For media consumption, the tablet supports Widevine L1, allowing for full HD playback on streaming services like Netflix. The stereo speakers are clear, though they lack deep bass, and there is no dedicated headphone jack.

Gaming performance is respectable for this hardware class. During my time with Fortnite at native resolution and low settings, I averaged between 45 and 50 frames per second. I also tested emulation; some GameCube titles ran at full speed, while PlayStation 2 emulation stayed just under full speed at around 50 frames per second depending on the game. The 3DMark Wildlife benchmark returned a score of 3,270, placing it in a similar performance bracket to the Nvidia Shield TV.

Battery life is advertised at 15 hours, which seems accurate for light tasks like reading or emailing at moderate brightness and 60Hz. Intensive tasks like gaming will reduce that runtime significantly but still provides multiple hours. It is worth noting that while a charging cable is included in the box, a power adapter is not. Additionally, this model lacks a GPS radio, which may be a consideration for users who rely on offline navigation.

Walmart typically maintains these devices on the market for about two years, and while it carries a one-year warranty, there is no formal commitment regarding future Android OS or security updates. This tablet represents a specific segment of the market where hardware value is prioritized over long-term software support.

For those looking for a large-screen Android experience without a high price tag, the Onn Pro provides a functional set of features that generally exceed the typical expectations for the budget category. It will be interesting to see if the video input functionality is refined over time, as that remains one of the more unique additions to an otherwise straightforward hardware package.