809 karma · joined September 11, 2022
that said, I'll probably never buy a 2-in-1 like this again. my primary use case for the surface was diagramming, note taking, and brainstorming. I used the pen and kick stand all the time. if I needed the keyboard I could pull it out and clip it on, tho after a while I rarely felt the need to bring it with me. by contrast the asus's flip-over mechanism feels too unwieldy and cumbersome. it can't be used at a kickstand type angle since the hinge isn't stuff enough to hold it up while writing on it. it's well-built, just the wrong use case.
Maybe I'm misunderstanding but it seems like they still depend on an attention based model to train their model? While it's interesting and would definitely enable the personalized AI they seem to be going for, I don't really see how they can say that it's not based on attention architecture. Someone still needs to shell out the big bucks to train that teacher model, no?
In this case, you're building what amounts to a proof of concept or a minimum viable product. Your constraints are mainly time and reliability, ie: it has to work and it has to be delivered by a certain time, so I'd argue the biggest factor in stack selection is your familiarity with the stack in question.
Besides that though, language will have almost no effective impact on your real world performance, which will be dominated by network latency.
You can test this for yourself to confirm: Write a script which sends random strings to a backend which simply writes the string to a database. Have the script print a timestamp before and after calling the API. Likewise have the backend print timestamps before and after each db call. Start the backend up on a small cloud VM, the database on another, and run your script from a laptop on Starbucks WiFi.
I can tell you right now that the laptop to backend latency will be around 50ms, while the backend to db latency will be around 5ms. Could you reduce that 5ms by a factor of 10 with the right language and db choice? Absolutely, but it won't make a difference to the end user.
The scale of fibre is the problem. Fibre makes sense when you're serving places where a single neighbourhood rivals the population of entire rural towns. They're close to backhauls, technicians, and enough customers to justify the cost. Which is why companies have dragged their feet on it in rural areas where the nearest full-time technician is a half-day's drive away. I mean shit, my mother lives less than an hour outside the capital city of my home province and can't even get DSL. It's too much copper for too few people to justify maintaining it at a price people are able to pay.
I've been using Starlink in rural Canada, prairies and mountains, where most people are either using Starlink, horrible past-gen satellite links, or microwave LOS systems. My father's been using a microwave LOS system at his farm for a decade.
All of them are affected by weather to some extent, but Starlink is the least affected, and no Starlink user I've talked to has been affected by congestion issues. Yes it's absolutely slower than fibre, it's slower than the DOCSIS 3 cable internet I was using before this. Microwave is usually cheaper, but Starlink is much faster (unless you're wealthy enough to get a beefy uplink). And importantly, Starlink ready to go right now.
Rural users often have to choose between nothing and a very expensive something. Fibre would be nice but wiring up every rural customer to fibre is a ridiculously time-consuming and expensive proposition. On the off-chance that a rural town grows instead of dies, then fibre would be better for them but at that point the economics of running fibre would make sense without the need for subsidies.
Rural people need alternatives to Starlink but fibre isn't it.
The keyboard itself was not good for a bunch of reasons, but the idea was gold. Individual, mechanical keys which could change their legends to suit the current context. You wouldn't have to memorize every possible layout before using it, and you could change the layout to suit whatever you're currently doing.
The closest equivalent I've seen would be the pre-mouse-era keyboards which could accept laminated paper legends for specific applications. The next closest, tho in the opposite direction, would be modern configurable keyboards with support for multiple layers layers.
If you zoom in it looks like the print head has two extruders but that's pretty meaningless without details.
We trialed it at my last company. It managed to impress the CEO and my security team at the same time for vastly different reasons.
if the F-35 program is cancelled again, textbook authors won't ever need to drum up another example.
As the author notes, doing this with rootless podman is a pain in the ass. If you want the containers to start with the system, then you have to have a system target which a user service waits for. The user service then has to reference each of the container quadlets in order to start them. It's fiddly and adds new failure points.
Even if you're not using rootless, the quadlets have to be run through a script which translates them systemd units, so it's awkward to use with declarative config management. Quadlet syntax and systemd syntax are also similar enough to be confusing at times.
It might be fine if you're building host images (eg: immutable dev env VMs). I won't be using this in production though, whether that's at work or in the homelab.
As an early adopter of digital cameras and convert to film photography, I agree... to a point. Producing creative work can be entirely about the output and those outputs can be as Fine as any Fine Art, not "merely entertainment".
Take writers for example. As a business and as a creative pursuit, writers have (on the whole) enthusiastically adopted every technology that's come their way. In large part that's because it helped them stop thinking about certain critical operations. eg: dealing with ink and nibs and loose papers. And while I'd imagine most writers derive a lot of fulfillment from the process, none of that compares to the feeling they get when others read and enjoy their work. Do some writers still use pen and paper? Absolutely, but does that mean their art is better or more Fine or more authentic?
Those analog tools might still be around, but it's largely because there are people who still enjoy those mediums, not because their modern alternatives are inherently Less in some way. If they were then all the paintings in your local gallery would have been done in tempera.
at my last place we relied heavily on APM traces in prod and kept the sampling rate relatively low. you can add metadata and send custom trace entries to take the place of most log entries. likewise the most valuable metrics could be derived from APM data easily enough. we kept shipping the must-have logs and custom metrics but their volume was surprisingly low once the noise was gone.
also, I don't think I'd ever let datadog slurp everything from AWS cloudwatch again. the amount of pure useless noise there far outweighs the value. a few targeted collectors from services like WAF and AWS SSO were all we really needed in the end.
TL;DR: be ruthless and cut cut cut. you might be surprised at just how much is never used. also, avoid letting datadog integrate with lambda or ECS, since those come with a per-unit cost which is fairly extravagant.
This could be useful if it can reliably get the core requirements correct though. I can see myself generating some ideas with this and taking them to an artist as a basis for them to start with.
OP: It would probably be good to have a "these aren't even close, please try again" button which allows for a couple free retries.
I hate it when my GPU decides to waste time on puzzles instead of rendering things.
spoiler: $1.9M to $3.5M used. less if you're down for a timeshare deal.
considering that it was rented in Colorado, parked right in front of the trump hotel lobby with the trump hotel sign right above it, and exploded right after, the simplest explanation seems to be that it was intentional and that causing a scene was more important to them than killing.
when you wind the watch, the mechanical energy is transferred to these springs.
pushing the illumination button allows the springs to release their tension into the generator, which powers the LEDs.
spoiler: a cool half-million