macNchz

It's surprisingly good! I wouldn't want to type on it all day long, but it works pretty well for me. I just did a cold typing test, not having typed on it in a few days, and got 55wpm touch typing the normal way, and 40wpm with thumbs. On a regular keyboard I can type >100wpm, so it's clearly not quite there, but I much prefer it to a smartphone keyboard, and I can be much more accurate on it without any sort of autocorrect.

Hiring a housekeeper to come every couple of weeks has pretty much directly bought me time, at a pretty reasonable price. I like living in a neat and tidy home, but never cared much for scrubbing grout or polishing the stovetop in my free hours. I’m delighted every time she comes, and I never wake up Saturday thinking I’ll have to vacuum under the couch cushions.

I recently bought a GPD MicroPC 2, a 7” laptop with a real keyboard. It runs Linux just fine, and it has been a fun experience of having a “real” computer with me much more often than I otherwise would. My version of programming on the subway has been programming on a park bench—it fits in a jacket pocket, or even the back pocket of some of my pants. The keyboard is tiny but easy enough to use with thumbs, or, with some practice, two-handed touch typing on a flat surface.

It’s nice to be less tethered to a desk, while also not having to carry a backpack and heavier full laptop, but still able to remote in and do what I need to do. I really enjoy having a fully capable Linux PC in my pocket vs a smartphone.

New York actually does have a points system, but since they're tied to the driver's license rather than the car itself, you only get them if you're actually pulled over, not from cameras. Within NYC there's a fair amount of camera enforcement, but comparatively very little by the police directly, so drivers whose licenses might otherwise be suspended via points are still driving around.

The mechanisms for keeping people off the road are also just weaker in the US—I believe the penalties for driving with a suspended license are comparatively lighter, plus if your license is suspended you can often still get a "restricted" license that still lets you drive to work.

I have noticed this going between Switzerland and Italy in particular—all of the cars going incredibly fast on the autostrada seem to have Swiss plates!

When New York State authorized the NYC speed camera program they explicitly precluded it from reporting to insurance, and made it not part of the “points” system that triggers license suspension if you accumulate too many infractions, so all that happens is that you get a $50 ticket each time.

If you don’t pay the tickets, your car is at risk of being booted, but if you don’t park on the street or choose to obscure your license plate when you do (how did that leaf get stuck there!?), there aren’t many repercussions.

There was an attempt at a program to actually seize these cars, originally it would have kicked in at 5 tickets/year for immediate towing, but it was watered down to 15 tickets a year triggering a required safe driving class. They sort of half-assed the execution of that, then pointed at the limited results and cancelled it altogether. There’s an effort to pass a state law about this, we’ll see if it makes progress.

It’s a very widespread problem, I think, and probably has a complex mix of causes, but my perception as a NYC runner, cyclist, and driver is that there’s a fairly small percentage of extremely antisocial drivers who we allow to behave badly with relative impunity, which itself moves the Overton window of driving behavior towards aggression/chaos, so to speak.

Very frequently when there is a newsmaking incident in which a driver runs people over in some egregious fashion, it turns out that they got dozens of speed camera tickets per year. We know who these people are, we just don’t seem to have any motivation to actually do anything about it.

The city has published research on this, showing drivers who get 30+ speed camera tickets in a year are 50x as likely to be involved in crashes with serious injuries or death, but efforts to actually do something about their behavior are consistently stalled or watered down. Other research points to various causes, including backed up courts and decreased enforcement generally.

https://www.nyc.gov/html/dot/html/pr2025/nyc-dot-advocate-fo...

https://www.nyc.gov/html/dot/downloads/pdf/nyc-driver-behavi...

There’s nothing super special about it, it’s just handy if you have some instructions that you don’t need the AI to see all the time, but that you’d like it to have available for specific things.

Maybe you have a custom auth backend that needs an annoying local proxy setup before it can be tested—you don’t need all of those instructions in the primary agents.md bloating the context on every request, a skill would let you separate them so they’re only accessed when needed.

Or if you have a complex testing setup and a multi-step process for generating realistic fixtures and mocks: the AI maybe only needs some basic instructions on how to run the tests 90% of the time, but when it’s time to make significant changes it needs info about your whole workflow and philosophy.

I have a django project with some hardcoded constants that I source from various third party sites, which need to be updated periodically. Originally that meant sitting down and visiting a few websites and copy pasting identifiers from them. As AI got better web search I was able to put together a prompt that did pretty well at compiling them. With a skill I can have the AI find the updated info, update the code itself, and provide it some little test scripts to validate it did everything right.

I also have a 980 that's going strong. A number of parts replaced over time, so refreshingly repairable. My mother in law has had a whole lineup of Chinese robot vacuums in the time we've owned Roomba and they haven't seemed miraculously more effective. Ours does get stuck occasionally, but it runs every day so...it's not a huge deal if it doesn't finish on any given day.

I do occasionally wonder how different things would be if JavaScript had come with a very robust standard library from early on.

I think it's a significantly lower barrier than employing people to watch feeds non stop or review every instance of motion or person-detection. It would be pretty straightforward for the camera maker to test and evaluate a handful of presets that ship with the cameras, and the current state of vision models is already pretty excellent at identifying things in a nuanced and flexible way in images and video.

Google has great distribution to be able to just put Gemini in front of people who are already using their many other popular services. ChatGPT definitely came out of the gate with a big lead on name recognition, but I have been surprised to hear various non-techy friends talking about using Gemini recently, I think for many of them just because they have access at work through their Workspace accounts.

All hype and thought experiments about superintelligence and open questions about creativity and learning and IP aside, this is the area that gives me the biggest pause.

We've effectively created a panopticon in recent years—there are cameras absolutely everywhere. Despite that, though, the effort to actually do something with all of those feeds has provided a sort of natural barrier to overreach: it'd be effectively impossible to have people constantly watching all of the millions of camera feeds available in a modern city and flagging things, but AI certainly could.

Right now the compute for that is a barrier, but it would surprise me if we don't see cameras (which currently offer a variety of fairly basic computer vision "AI" alerting features for motion and object detection) coming with free-text prompts to trigger alerts. "Alert me if you see a red Nissan drive past the house.", "Alert me if you see a neighbor letting his dog poop in my yard.", "Alert the police if you see crime taking place [default on, opt out required]."

I did have Prime for like 10 years, I just eventually realized that between the not-infrequent annoyances with shipping and the endless search results full of total junk and/or fake products it wasn’t as convenient as I’d thought, it more so just became my default.

There’s a corner store about a two minute walk from my front door, I’m certain their toilet paper is more expensive than Amazon’s, but I can have it right now if I want, and I’m not dealing with the stupid interface asking me if I want “18-count (345 sheet, 9 pack)” or the “XL 27 count (256 sheet, 5 pack)” version of the same product.

I disagree with this article—its premise relies too heavily on the oft repeated, oft misunderstood line “there is no security in obscurity.”

This concept is used to argue that obscurity shouldn’t be used at all as a defense mechanism, when really all it means is it shouldn’t be your only line of defense.

Obscuring aspects of a system can contribute to its overall functioning: it’s a filter for the laziest of adversaries, and it creates an imperative for more motivated ones to probe and explore to understand the obfuscation, creating signal and therefore opportunities to notice their behavior and intervene.

I think for anyone who has dealt firsthand with mitigating online fraud, hackers, spam, trolls, cheating etc, the idea of having completely transparent defense mechanisms is pretty much ludicrous.

People in my circles in the US (in an area with tons of alternative options) look at me like I have two heads when I say we don’t have Prime and never shop on Amazon. For many, I think, Amazon has simply been the default option to buy anything for long enough now that it’s ingrained muscle memory.

A lot of my notes and tasks wind up having bits of code and sometimes large data files associated with them, so I've landed on a similar path of using plain text/org mode files, but aided by a little shell function `today` that creates-if-not-exists a new subdirectory named for the date whenever I use it:

    function today() {
        TODAY_DIR="$HOME/today/"
        DATE_DIR=$(date +'%Y-%m-%d')


        if [ ! -d  $TODAY_DIR$DATE_DIR ];
        then
            mkdir -p $TODAY_DIR$DATE_DIR
        fi;

        echo $TODAY_DIR$DATE_DIR
    }

So I just do something like `emacs $(today)/tasks.org`. Easy to grep across time, copy things forward (I guess I could do with having `yesterday` and `tomorrow` as well). It's really nice to just use basic CLI tools and little scripts to manage notes and todo lists. Project specific stuff gets a subfolder name every day so it's easy enough to glob ~/today/*/{project}/....

It's a sort of landing zone for all of the miscellaneous artifacts I might deal with on a given day as well:, e.g. `wget -P $(today) https://site.net/cooldata.gzip`.

I always figured it was an engagement optimization thing—there were people mass producing content using popular characters and just throwing tons of stuff at the wall, and the ones that veered unsettling/bizarre wound up getting lots of engagement so they kept doubling down on it. That kind of feedback loop is certainly responsible for many other curious traits of online content that is circulated in algorithmically-curated feeds.

The tighter the loop between content creation (e.g. when you can generate unlimited content essentially for free) and the ability to measure its success (engagement), the more social media becomes a sort of genetic algorithm for optimizing content to be the most addictive possible at the expense of any other attribute.

The notable source of bad tailpipe emissions in NYC are heavy diesel trucks, which, to my understanding, produce a large proportion of tailpipe particulates (and NOx) in the US, despite being a small fraction of overall vehicles on the road. There are strong correlations with heavy truck traffic and asthma rates.

If you're interested in much faster LLM coding, GLM 4.6 on Cerebras is pretty mind blowing. It's not quite as smart as the latest Claude and Gemini, but it generates code so fast it's kind of comical if you're used to the other models. Good with Aider since you can keep it on a tighter leash than with a fully agentic tool.

An interesting test in this vein that I read about in a comment on here is generating a 13 hour clock—I tried just about every prompting trick and clever strategy I could come up with across many image models with no success. I think there's so much training data of 12 hour clocks that just clobbers the instructions entirely. It'll make a regular clock that skips from 11 to 13, or a regular clock with a plaque saying "13 hour clock" underneath, but I haven't gotten an actual 13 hour clock yet.

Those do exist, they’re called cell signal boosters. Once upon a time, I believe, some American cell providers would give you one for free if you had bad signal at home, which mattered a lot more before phones all had wifi calling.

What is the actual return on that investment, though? This is self indulgence justified as « investment ». I built a pretty beefy PC in 2020 and have made a couple of upgrades since (Ryzen 5950x, 64GB RAM, Radeon 6900XT, a few TB of NVMe) for like $2k all-in. Less than $40/month over that time. It was game changing upgrade from an aging laptop for my purposes of being able to run multiple VMs and a complex dev environment, but I really don’t know what I would have gotten out of replacing it every year since. It’s still blazing fast.

Even recreating it entirely with newer parts every single year would have cost less than $250/mo. Honestly it would probably be negative ROI just dealing with the logistics of replacing it that many times.

There’s a difference between something that has existed for a few years that lots of people have heard of, and something that people have been buying their entire lives, and that their grandparents also bought for their entire lives. As to distribution—the internet certainly makes it logistically easier to get your product to consumers, but an infinitely large store shelf still means you’re competing for consumer attention, and the big players already have that attention for their existing successful products.

I think it’s more reasonable to consider Coca Cola as having a significant brand value moat, given that they’re 140 years old and one of the most recognizable brands in the world. That also gets at the other side of their moat: distribution. Coca Cola is available basically everywhere, and a challenger would have to invest massively to simply get in front of as many people on shelves. In that way, other companies (Google, Microsoft, Meta) still have significant legs up on OpenAI. Way too much in play right now to declare any winners.

It’s how they do it in the Mixpanel setup guide: https://docs.mixpanel.com/docs/quickstart/identify-users#cod...

Also probably people on the product marketing team want to have identifying info in their dashboards of top users and churn risks and whatever, and someone has to be the one to tell them no.

This is a strange feature of hospitals that seems very much like we should know better, yet it’s pervasive. Hospital-acquired delirium is very common and awful, and is associated with bad outcomes after leaving the hospital—higher risk of death, and ongoing dementia-like problems.

We know sleep is so critically important to health, yet everything about most hospitals seems basically incompatible with proper sleep: beeping, lights, middle-of-the-night blood draws, shared rooms, no sense of day or night. Not dissimilar to how how we know that people make terrible decisions when they’re sleep deprived, but continue to have resident doctors work unreasonable hours, this is a status quo bias that harms people.

The most effective AI coding assistant winds up being a complex interplay between the editor tooling, the language and frameworks being used, and the person driving. I think it’s worth experimenting. Just this afternoon Gemini 3 via the Gemini CLI fixed a whole slate of bugs that Claude Code simply could not, basically in one shot.

I see a bunch under major SaaS vendor namespaces that have millions of weekly downloads…?

Ground meat is different from mechanically separated meat, which is a paste. At least personally I haven’t seen ground meat for sale in the US that lists MSM as an ingredient, though it may exist.