The scene desperately needs some clipping on the boundaries. If you use an app like Scaniverse, you can add a bounding box to cull far away points which are often poorly reconstructed.

If you have a newer iPhone with a LIDAR scanner, highly recommended. You can make dolls-house renderings of your house or garden which is surprisingly useful for planning and measuring walls/features.

By far the best AI+human customer support mechanism I've experienced is through SMS/messages. They support auth, they're asynchronous, there's no app or custom interface to timeout, it's easy to send complex queries as text and you have the log right there. Apple does this really well. Delta also does, surprisingly, because their AI phone bot is garbage. It's also presumably easier for the human agents to multi-task.

This is usually the same the other way; many vendors will give business clients net 30. That's nice if you're a small company and need to plan ahead. But occasionally, because you're considered small (liability), some vendors will want the money up-front. So unless you're very careful with cashflow, you end up in situations where your main sources of income (big contracts) are coming in after you need to spend money on a widget to fulfill deliverables.

Depending on the situation, the contract can demand the client purchase/ship things and work doesn't start until you have them in hand. This is usually the best route as you now have an out, and it's not an unreasonable request, but it doesn't always pan out that way!

You might execute formally verified code within an RTOS, which is your two worlds? Consider you have some critical control loop, like an autopilot (see Ardupilot). That control loop must run at some minimum rate, and the action of the system must be well characterized. Similarly you might want to guarantee that you sample a bunch of sensors frequently enough (so the most recent reading is no older than some time period).

I've switched to UTZ Rako/Euroboxes for longer term storage. I even bought a beat-up dolly so I could easily transport 60L boxes around. They stack, they're divisible (e.g. 2x30L on top of a 60L) and the smaller ones fit neatly into a KALLAX cubby. You can buy them used for cheap, if you're willing to spend an afternoon scrubbing factory dirt off them. But they're not significantly pricier than Really Useful.

There are other suppliers like Auer, who make all kinds of interesting variations like toolboxes and latching/lockable boxes, but can get really expensive. You can get insert containers for them, but same problem: no transparent lids, only generic gray unless you want to buy 100.

I've been lusting after some of the Sortimo boxes that Adam Savage recommends, but I can't justify 50-100 quid per compartment box.

As for the original article... I like the idea of dots, but I would try a gridded label with sharpie marks. Having worked in a lot of workshops and labs though, boxes are not efficient. You want a good rack/drawers for things you use all the time (tape). I do like one box per project for convenience, which is often more useful than a box with generic grouped parts. If you really need to, you can do things like cut SMD tapes for each project. This way is much easier to drop back into something you only have time for on the weekend, and it's also what we would do in hardware shops (single sorted organizer with the BOM items for a project).

I do agree about the hardware side being surprising. When I was working on electronics for work, having a 4 channel scope was indispensable. But most of the time, debugging on chip/breakpoints are enough. I switched to a 4 channel mixed signal Picoscope.

(Someone else mentioned kitchen containers. I spent some time in a professional kitchen which hooked me on Cambro-style containers, or whatever Nisbet's sell in the UK. Also standardized alu sheet pans and matching silicone mats for baking.)

Downside is no backlight which many users tout as an improvement, or praise it as a minimalist perk. I don't really agree, but it does mean that the ink surface is closer to the pen so there's less parallax error. It makes it less usable as an ebook reader though, for example on a flight you'd have to use the blinding overhead lights.

Sure the price is comparable to 20+ notebooks. I think if you actually use notebooks, they're good. If you don't, it's questionable whether it'll change your habit. It also doesn't replace the satisfaction of a nice ink pen on nice paper. I have a collection of fountain pen ink that I've used since university (for years of daily lecture notes which is more writing than I'm ever likely to do again - we're talking up to 20 A4 sides a day) and the bottles are still practically full. So good writing equipment can be very economical. There are other issues like no colour (on mine) and PDF support is still ropey.

I keep a few converters for older devices and servers that don't have (m)any C ports, but as far as a consumer "forever cable" goes, TB5 feels close. Certainly the cable's bandwidth is beyond what most people need, unless you're editing 8k video or continually shuffling hundreds of GBs between external disks.

What I found funny was that if you asked ChatGPT to provide a score recommendation, it was also significantly higher than what that reviewer put. They were lazy and gave a middle grade (borderline accept/reject). We were accepted with high scores from the other reviews, but it was a bit annoying that they seemingly didn't even interpret the output from the model.

The learning experience was this: be an honourable academic, but it's in your interest to run your paper through Claude or ChatGPT to see what they're likely to criticise. At the very least it's a free, maybe bad, review. But you will find human reviewers that make those mistakes, or misinterpret your results, so treat the output with the same degree of skepticism.

Thinking about it, we also had some "fancy" packet ramen from Momofuku. Good example there - those noodles take forever to cook compared to the deep fried ones. You'd have to soak, nuke in the microwave and still wait ages.

Most of the coffee we took down were light roast and how well the beans survived shipping/storage, how well they were roasted mattered much more.

There are a bunch of cafetieres as well, but I don't like the silt even with some of the techniques designed to minimize it.

We'd definitely have kidney beans in chili and some other dishes, but I got the impression it was a hassle otherwise.

> Re: coffee, mixing concentrated cold brew with hot water makes a pretty smooth cup

Friend and I ran a weekly pop-up espresso bar and did a lot of experimenting over the winter. The USAP "house" beans are quite dark, but at least they're roasted within a year or two because coffee is always available and we go through a lot of beans every season. Except the decaf. That stuff is decades old.

People often bring down a big bag from one of the roasters in Christchurch. We personally shipped down a lot of specialty coffee, mostly made V60 and aeropress. The outbuilding where our telescopes live also has a Chemex and an automatic.

We almost ran out of sugar in 2021 and Rothera sent us a bag of Tate and Lyle in break-glass-in-emerhency box on one of the early transit flights the following summer. That's still hanging in the galley. Cream also goes pretty quickly, and forget about eggs. But you only need "egg product" anyway.

The foods that tend to be avoided are pasta and beans, or really anything which has to be boiled. There's a massive pressure cooker but it's a pain to use and clean. It's also hard to brew coffee if you tend to use off-the-boil. The best you'll get is about 93 C. Espresso is fine as its pressurised anyway.

> Honestly? We should address X first. It's a genuine issue and we've found a real bug here.

Honorable mention: "no ". I guess this helps reassure adherence to the prompt? I see that one all the time in vibe coded PRs.

However, in some harnesses the model is given access to the old chat log/"memories", so you'd need a way to provide that. You could compromise by running /compact and pasting the output from your own summarizer (that you ran first, obviously).

Something I miss from the Windows side is sub-kg machines, at least since Apple discontinued the 12" Macbook. It makes a surprisingly big difference when traveling, especially with Asian carriers that have hard carry-on limits. The Thinkpad X1 Carbon is a fantastic form factor, though the older Intel chips run incredibly hot. I repurposed that as a garage/workshop Linux machine. Unfortunately, the price differences between Mac/Windows also disappear when you start looking at those higher-end machines.

Your main concern is to stay dry and minimize snow incursion. Either wear ski pants that act as gaiters, use gaiters or use boots and socks that are high enough that you won't get snow down the sides.

If you buy boots with insulation, try not to compress it. Otherwise be aware that if you don't keep moving, your boots will eventually cool to ambient and it's pretty hard to get that temperature back up.

Check grip? Hard to test but warm doesn't necessarily mean any good on slick ice. Spikes work well if you're going on a hike and there's a lot of packed snow mixed with ice.

Don't forget good socks. Doesn't need to be anything fancy, but wool is by far the best material (not necessarily merino as it tends to be too thin). You may need to size up because of the extra padding.

Also luxury, but fan assisted boot drying/warming stations are great. They make quite a big difference if you go out a lot because moisture build-up takes ages to dry otherwise.

It's worth checking different versions of Claude Code, and updating your tools if you don't do it automatically. Also run the same prompts through VS Code, Cursor, Claude Code in terminal, etc. You can get very different model responses based on the system prompt, what context is passed via the harness, how the rules are loaded and all sorts of minor tweaks.

If you make raw API calls and see behavioural changes over time, that would be another concern.

Blurring is a convolution or filter operation. You take a small patch of image (5x5 pixels) and you convolve it with another fixed matrix, called a kernel. Convolution says multiply element-wise and sum. You replace the center pixel with the result.

https://en.wikipedia.org/wiki/Box_blur is the simplest kernel - all ones, and divide by the kernel size. Every pixel becomes the average of itself and its neighbors, which looks blurry. Gaussian blur is calculated in an identical way, but the matrix elements follow the "height" of a 2D Gaussian with some amplitude. It results in a bit more smoothing as farther pixels have less influence. Bigger the kernel, more blurrier the result.There are a lot of these basic operations:

https://en.wikipedia.org/wiki/Kernel_(image_processing)

If you see "Gaussian", it implies the distribution is used somewhere in the process, but splatting and image kernels are very different operations.

For what it's worth I don't think the Wikipedia article on Gaussian Blur is particularly accessible.

These are different requirements. The issue I described was not a power outage and having a well managed UPS wouldn't have made a difference. Nothing shut down, but we lost 5G in the area and T-Mobile's modem is janky. My point is that it's another edge case that you need to consider when self hosting, because all the remote management and PDUs in the world can't save you if you can't log into the system.

Of course there's all you need is a smart plug and a script/Home Assistant routine which pings every now and again. There are enterprise versions of this, but simple and cheap works for me.