joshvm

Any expat who has lived in Europe knows the pain of having to run to the train station on a Sunday because you ran out of some ingredient.

The simplest option is to print the part raised at an angle so the layer lines aren't parallel to faces. Clough42 has some good videos on support/rib design in Fusion: https://www.youtube.com/watch?v=XXaLxSmtnbQ, based on https://www.youtube.com/watch?v=8NKVNwVaZU0.

But you can definitely get printers to dump a blob of filament out without worrying about cooling problems, if the extruder speed is high enough. I was debugging some issues in P2PP (a post processor for the Mosaic Palette. One problem was that the printer would extrude all the filament at the start of some travels instead of along the path.

When I was younger, I used to think the Shard was daylight robbery. But 16 quid for a signature cocktail and a view (or 12 for a spirit + mixer) is no longer outrageously expensive in London and you'd pay that to go up a skyscraper in many countries anyway. Both Rockefeller and the Empire State are $40+, though you can go outside. Even Aqua Shard, which is apparently in the Michelin guide, "only" charges about 20. Which is bordering on outrageous, but the view is great.

I recently started playing around with my Palette 3 again (on a MK3.5S). It’s an amazing piece of engineering that has a reputation for being frustrating to use. It’s now discontinued but Mosaic still sell spare parts and it’s designed to be stripped down and repaired. Despite the problems, it was built by good engineers - it uses one torx head for all screws (and comes with a driver). The next printer I get will probably be an INDX system though, the future is multitool.

I had written it off, because of how irritating Mosaic’s cloud slicer is. I’ve been pleasantly surprised how well it works with a fork of P2PP (vhspace/p2pp), a post processor for PrusaSlicer that is completely local. All it does is swap out the filament change commands with Palette splice instructions. I fixed a few gcode interpreter bugs that solved the issues I had with bed calibration and extrusion, and even splicing seems quite reliable. I’ve been using it for simple 2-color same-brand prints that would require a lot of manual changing, so not complex but it’s much nicer to use than an MMU2.

I’m excited to try this out for sure.

That one has a strange rendering error for me, the trees and horizon are in front of the mill building and the exterior isn't properly rendered unless you are in orbit mode. But my mind was a little blown when i discovered that I could walk up the stairs. It needs shift to run!

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.

I don't know about useful, but the most visible one is copywriting. Even when there's a human involved, every startup/small org I know runs content through them. (And that includes this article.) It's definitely something that companies want even if they don't necessarily need it (like analytics).

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 the sort of economics that small companies face when working with large companies, particularly when physical things/CAPEX are involved. Large companies expect net 30/60 terms to pay you. That's much simpler for their accounting/purchasing department. This bureaucracy occasionally necessitates nudging, especially if the intermediary you're dealing with didn't set up the invoice request on time in whichever SAP/Salesforce/Oracle system they use.

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!

RTOS doesn't give any guarantees about "safe to compute/execute" - that's more the domain of formal verification. In the sense that you can make guarantees about how the program will behave given some domain of inputs. But predictable (or bounded) latency, yes.

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).

Really Useful Boxes are excellent. The best feature for me is that the lids are raised, so you have some room that's slightly higher than the rim of the crate. Sure, buy a bigger crate, but it's nice to pack things just about level and not worry about one annoying part poking up.

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.)

I have a Supernote which I like primarily because it's repairable and the developers are very responsive. It's the A5 version. It's very nice to write on and if you haven't tried eink in a while, it's pretty impressive. The soft surface is also a replaceable film. It has a Lamy colab pen which is very nice.

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 started buying Belkin TB5 cables which are around $50 a pop. They can easily power a laptop at full load and can stream video at any reasonable resolution/framerate I might need. I've yet to find a need for an NVMe faster than 20 GBps nevermind finding USB4 enclosures, or that the cable supports up to 80. They're also not nearly as chunky as the Dell cables, which are good, but seem to have very rigid shielding.

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.

https://www.itsquieterfilm.com/trailer is the official site.

This definitely happened to a paper that I submitted a couple of years ago. ChatGPT 4 was the frontier. The reviewer gave a positive, if bland, summary with some reasonable suggestions for improvement and some nitpicks. There were no grammar or line-number comments like those from other reviewers. They were all issues that would have been resolved by reading the appendices, but the reviewer hadn't uploaded into ChatGPT. Later on I was able to replicate the output almost exactly myself.

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.

This is true, and even with black tea where you'd normally want hotter, I don't think anyone really pays attention

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.

Do they not do soaked beans? Leave them in water for 2 days and they shouldn't need a full boil I wouldn't think?

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.

The few times I've baked there, it's been a pretty good experience. There's a full height proving cabinet, yeast works really well at altitude, the ovens have steam injectors, there are good mixers, a commercial fryer. In many ways much easier than baking at home, but probably not a patch on a good bakery.

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.

No mention of Claude/ChatGPT's favourite new word genuine and friends? They also like using real and honest when giving advice. Far as I can tell this is a new-ish change.

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

Honorable mention: "no <thing you told me not to do>". I guess this helps reassure adherence to the prompt? I see that one all the time in vibe coded PRs.

Have you tried writing that as a skill? Compaction is just a prompt with a convenient UI to keep you in the same tab. There's no reason you can't ask the model to do that yourself and start a new conversation. You can look up Claude's /compact definition, for reference.

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).

The same is also true within the Macbook line. The 14" Pro is smaller and nearly 2lbs lighter than the first 13" unibodies. I have my 2009 college laptop on a shelf as a memento and it feels pretty chunky. This hasn't changed much in the M-series though, and the M5 is slightly heavier than the M1.

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.

Baffin make some of the best cold weather boots. We use them in Antarctica, though you probably don't want the chonky -70C rated ones. I have some lighter boots rated for about -40 and they're great. Really any good gore tex mid-ankle hiking boot is probably fine. Whether you need cold rated boots is going to depend on where you're walking.

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.

My gut feeling is that performance is more heavily affected by harnesses which get updated frequently. This would explain why people feel that Claude is sometimes more stupid - that's actually accurate phrasing, because Sonnet is probably unchanged. Unless Anthropic also makes small A/B adjustments to weights and technically claims they don't do dynamic degradation/quantization based on load. Either way, both affect the quality of your responses.

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.

Reference managers have existed for decades now and they work deterministically. I paid for one when writing my doctoral thesis because it would have been horrific to do by hand. Any of the major tools like Zotero or Mendeley (I used Papers) will export a bibtex file for you, and they will accept a RIS or similar format that most journals export.

I am vaguely aware of stuff like Gaussian blur on Photoshop. But I never really knew what it does.

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.

its not really about uptime for the homelab, its about graceful shutdown/restart.

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.

My home server doesn't need to be high availability, and the BIOS is set to whatever state prior to power loss. I don't have a UPS. However, we were recently hit with a telco outage while visiting family out of town. As far as I can tell there wasn't a power outage, but it took a hard reboot of the modem to get connectivity back. Frustrating because it meant no checking home automation/security and of course no access to the servers. I'm not at a point where my homelab is important enough that I would invest in a redundant WAN though.

I've also worked in environments where the most pragmatic solution was to issue a reboot periodically and accept the minute or two of (external) downtime. Our problem is probably down to T-Mobile's lousy consumer hardware.

I've found that for any sort of reasonable task, the free models are garbage and the low-tier paid models aren't much better. I'm not talking about coding, just general "help me" usage. It makes me very wary of using these models for anything that I don't fully understand, because I continually get easily falsifiable hallucinations.

Today, I asked Gemini 3 to find me a power supply with some spec; AC/DC +/- 15V/3A. It did a good job of spec extraction from the PDF datasheets I provided, including looking up how the device performance would degrade using a linear vs switch-mode PSU. But then it comes back with two models from Traco that don't exist, including broken URLs to Mouser. It did suggest running two Meanwell power supplies in series (valid), but 2/3 suggestions were BS. This sort of failure is particularly frustrating because it should be easy and the outputs are also very easy to test against.

Perhaps this is where you need a second agent to verify and report back, so a human doesn't waste the time?

AvE’s Juicero teardown is also a classic:

https://youtu.be/_Cp-BGQfpHQ?si=V-LLVn1cQmrMsiDZ

Also check if you're in education at any level. Most university libraries subscribe to what used to be Safari and you can SSO the full (enormous) catalogue. I didn't realize this for quite a long time as it's not widely advertised. There are ton of books that aren't the traditional animal-drawing tech titles, including Manning, as well as some lecture series.

But the app is pretty kludgey and it's way more locked down than other publishers who will give you chapter PDFs.

At least it's a good way to skim books to see if they're worth buying a physical copy.

I would also split the concerns:

1. Compliancy with relevant standards. HIPAA, GDPR, ISO, military, legal, etc. Realistically you're going to outsource this or hire someone who knows how to build it, and then you're going to pay an agency to confirm that you're compliant. You also need to consider whether the incumbent solution is a trust-based solution, like the old "nobody gets fired for buying Intel".

2. Domain expertise is always easier if you have a domain expert. Big companies also outsource market research. They'll go to a firm like GLG, pay for some expert's time or commission a survey.

It seems like table stakes to do some basic research on your own to see what software (or solutions) exist and why everyone uses them, and why competitors failed. That should cost you nothing but time, and maybe expense if you buy some software. In a lot of fields even browsing some forums or Reddit is enough. The difference is if you have a working product that's generic enough to be useful to other domains, but you're not sure. Then you might be able to arrange some sort of quid pro quo like a trial where the partner gets to keep some output/analysis, and you get some real-world testing and feedback.

It seems like a mixture of heuristics, explicit filtering and user reports.

https://help.kagi.com/kagi/features/slopstop.html

That's specifically for AI generated content, but there are other indicators like how many affiliate links are on the page and how many other users have downvoted the site in their results. The other aspect is network effect, in that everyone tunes their sites to rank highly on Google. That's presumably less effective on other indices?