Have you noticed that some of your custom Windows icons have reverted to the default look after the latest update? If so, don't worry; you're not alone. Microsoft has made some tweaks to how custom icons are handled with Windows 11, and if the operating system cannot verify that they came from a good source, it will silently change them back to defaults without warning you. Yes, it's one of those rare moments where what may seem like a Windows bug is, in fact, entirely by design.

People have now spent enough time with Forza Horizon 6 to gauge how good a game it is, and so far, the reviews have been shining. However, that doesn't mean that Playground Games can rest easy. The development team has announced that it has noticed a strange save game wiping bug that affects both Xbox and PC gamers, and while it has made recommendations on what gamers can do to keep their saves intact, it sounds like it doesn't quite know what's wrong yet.

If you've been hanging out in the self-hosting circles lately, everyone seems to declare that a powerful local AI setup is far better than paying for a cloud LLM subscription forever. The reasoning is that instead of paying $20 perpetually for Claude Pro (or a service of your choice), you could absorb a one-time investment in a beefy GPU, and enjoy all the inferencing you want on your PC. Eventually, the whole thing is sure to pay for itself, right? Well, it turns out that the answer isn't as simple. Whether a local setup ends up being better for you depends on your use case, preferences, and how much time you spend on LLMs daily. There are also hidden costs with both cloud and local models that many people don't consider. You'll have to contend with trade-offs either way. What will determine the winner for you, specifically, is which trade-offs you're more comfortable with. I did the math, which is the easy part; what's harder is deciding what level of investment, quality, control, and privacy you're after.

‘Listen, that's not what I'm here for.' | Image: Apple Our early testing has already shown that Siri AI knows when to shut up, and that's very much by design. In an interview with Mostly Human spotted by MacRumors, Craig Federighi said Apple's new Siri won't act all sycophantic like chatbots made by OpenAI, Google, and others."As you may know, if you use many of the existing chatbots, they're really focused on engagement to a large degree," said Federighi, who is responsible for software at Apple. "And sycophancy, right? They kind of want to pull you in. They might encourage you to reveal things about yourself, and then use that as a basis to establish a connection."Apple purposely took a di …Read the full story at The Verge.

Jellyfin is host to dozens of plugins, but they aren't all created equal. Some plugins serve a decidedly specific purpose, while others were originally designed (and later shared) by a single user seeking a solution to a problem. I've tested a lot of these plugins, but if I had to get rid of them all, there's only a single one I'd truly fight to keep: Jellyfin Enhanced. This plugin is kind of like a Swiss Army knife; it does so much, and is easily one of the most versatile plugins I've ever worked with.

We've seen plenty of retro consoles built using a Raspberry Pi or an ESP32 chip, but they often require an external controller to play. However, what if you could have a handheld console where the screen is on the controller itself? Or, better yet, if you had an ESP32 device that could act as a handheld or a controller? Well, someone has made just that, and you can help fund it right now.

Filament prices coming down should have felt like the big win this year. A spool that costs less is easy to understand, easy to justify, and easy to celebrate when the printer is already sitting there hungry for plastic. I’m not going to pretend I didn’t notice cheaper PLA and PETG when I was restocking. I absolutely did, because I’m still human and I still enjoy pretending a bulk filament order is a financially responsible act.

The funny thing about home lab hardware is that “too much” usually has a very short shelf life. I set up a LattePanda Sigma with 32GB of RAM and a 500GB NVMe SSD, expecting it to be excessive for most of what I do. On paper, it looked like the kind of board that should spend half its life waiting around for a workload worthy of it. In practice, it took almost no time at all before I started finding jobs for it.

Claude Code gets most of the attention these days, but Codex deserves a lot more credit than it gets. I've been using the $20 Pro plan for a while now, and one of the most useful workflows I've found is using it to test products. I enjoy building things, but once you've spent days or weeks working on the same project, it becomes difficult to see it through a new user's eyes.

Sooner or later, a NAS or home lab that's pulling double duty hits the same wall: too much data, not enough space to store it. You go to add one more drive, reach for a SATA cable, but find that there's not a single spot available, or you've had your attention caught by a used enterprise drive that has very compelling price-per-terabyte value. Either way, the solution is largely the same, and it's in the form of a PCIe add-in card called a Host Bus Adapter, or HBA. HBA cards aren't sexy, but they're incredibly useful cards that add a ton of potential capacity to your home lab or NAS, and they're worth exploring if you haven't already. As with any used enterprise gear, though, they come with their own unique edges that need some ironing out.

It's intended for camping, but BougeRV's T1 light is so versatile that it also makes for a great summer travel companion that'll continue to light the dark corners of your life long after you return. I used it for a few months during a recent road trip in my camper van, and it's the light I keep reaching for again and again now that I'm home.The telescoping light features three articulating LED arms that together direct up to 3,000 lumens of white, warm, or red light in any direction you want. It can illuminate an area over 1,000 square feet and function as a flashlight, a mood light, and a 57Wh USB-C power bank that can deliver up to 30W …Read the full story at The Verge.

Self-hosting LLMs on a regular PC always felt like a partial win to me. The hardware does what you ask of it most of the time, but you can also feel you're running mid-tier stuff on a gaming GPU and pretending it counts as a serious AI rig. Mine's been doing fine with 8GB for months at this point, but there are still limits that show up.