Good bread exists, it’s just not cheap like it is in Europe.
375 karma · joined April 24, 2025
Good bread exists, it’s just not cheap like it is in Europe.
You shouldn’t be giving take homes unless they’re either short, or the applicant passed a screen and you’re investing time. Otherwise how are you “scaling” the review? Claude? Hidden test suite (not bad)? Some sort of leaderboard (bad, rewards people with time), something else?
(Aside, for a lot of cooking I’ve gone back to dead trees and I realized one day I have so many cookbooks, why am I looking at some lifestyle recipe blogger?)
Or the information actually goes out of date and best practices change. There’s been like 25 years of standard revisions since I first learned C++.
For kids, it's 50 hours of road experience, and 10 at night, but a parent can sign that off. Maryland looks the same? You've got to do a knowledge test to get a learner's permit, 6 hours behind the wheel + 30 class, and then you get a provisional which can be upgraded to full. NY is also a 50/15 hour parent declaration and maybe(?) a supervised course for a few hours.
https://www.stateofflorida.com/drivers-license-steps/ https://dmv.ny.gov/driver-license/complete-pre-licensing-req...
Making right-on-red illegal wouldn't be bad either. The number of times I've almost been run over when a car is stopped in the middle/straight lane and blocks line of sight to a right-turning car that doesn't look.
DUIs are at least treated seriously. It's one of the few offenses that will get a visa instantly revoked. Same in Canada I think.
This is common in image generation pipelines because if you find an image you really like, you can store the seed and then reproduce it with small tweaks, otherwise - to quote Borges - “Look at it well. You will never see it again" User-facing deterministic pipelines do exist for generative AI.
I know you make this argument in your post, but that's really the answer if you want repeatable results. For a classifier or a detector, determinism is a requirement, but for an LLM non-determinism desirable property because it feels like a more natural conversation. The downside is it's extremely difficult to replicate a response without pointing the model to an earlier conversation.
And specifically for the RPG case, don't you want non-determinism? You don't want the model spinning up the same identical person if you say "Generate me an NPC character sheet for an innkeeper". This was a complaint that people had in the past, that models would regurgitate the same scenarios or the same jokes.
Where I suspect DMs run into trouble is not randomness, but lack of self-consistency in worldbuilding. Say you generate an NPC and then refer back to them later and the model gets some details wrong. You could compare to a system like Dwarf Fortress where everything down to the genealogy and faction relationships are rigidly generated.
While many camera sensors use MIPI/CSI, you need enough lanes to transfer the data, the driver support in the kernel and other pipeline bits to get good images from the bayer. Almost all “real” cameras use ASICs or FPGAs to clock out the images. Additionally sensor companies are miserable to deal with in small volume and datasheets are under NDA. You’re much better off buying a camera from a machine vision company over USB3 or Ethernet, but you need one which properly enumerates as a video device (many do not). You can still do nice stuff like hardware sync/trigger from the Pi.
You're better off setting a budget and buying the best machine you can afford in that range, or picking a VRAM target and accepting the class of models you can run on it. Those models will almost certainly improve over time and your skills will adapt to the limitations. Hardware is so valuable right now that it's not even likely to be a significant loss if you had to sell.
Right now I think 24 GB is probably the best bang for your buck (used 3090), because you also get a high end gaming/gpgpu device which is nice anyway. 32 GB you can do with AMD or Intel, but NVIDIA is megabucks and at this point you're really paying for RAM. Unfortunately the ship has sailed on "reasonably" priced RTX 6000s, which at one point were about $7k and are being listed at $10k++.
Sure, try not to be useless, but if the company doesn’t have guardrails that’s not on them. If an intern deletes something: why did they have access in the first place? Why wasn’t there a backup?
What's the comparison at $50-100?
> We’re starting by just giving you detailed body composition maps — and we’ll be submitting regular test results to the FDA for increased capabilities.
As far as I understand ultrasound there's no reason you couldn't do this, it's just infeasible to do a full body scan with a hand probe and you get covered in goop.
I could steer the conversation to explore if the candidate had insight into why GasTown or other "swarm" models exist though. Do they know what a system prompt is? How context is managed/affects output? How harnesses can delegate to simpler models, and when to use different classes of model? I wouldn't expect anything particularly deep here, but bonus points for a solid understanding/experience of the tooling and how to get the most out of token budgets is valuable.
To the parent point: even if you refuse to use LLMs for whatever reason, I would expect any coding applicant to have at least tried them. I'd assume the person is living under a rock if they said no.
https://wiki.openzim.org/wiki/Build_your_ZIM_file
EDIT: https://get.kiwix.org/en/solutions/applications/kiwix-reader...
People tend to assume the capex is thrown away but as we’ve seen with RAM, don’t be so sure you won’t be able flip it if you need to.
It’ll fix up environments, bump versions, fix compiler issues and outdated code, give you a better user interface (or create one to begin with). I usually point to this when people question if AI has productivity benefits, because it’s a massive time saver for me. I have spent hours in the past trying to chase down some obscure platform-specific GCC error on Stack Overflow.
Ideally pay it forward and PR the things you fixed.
Normally you’d shoot at 180 degree shutter angle (exposure time is half the frame time). This produces a “cinematic” blur that doesn’t look choppy, especially when projected at the same rate. So if you’re shooting 24 fps video, try shooting at 1/48. This is slow compared to most handheld still photography, which is why you start to need ND filters on cine cameras, especially if you also want to shoot wide open.
Stutter is particularly noticeable for fast panning landscapes where there’s uniform motion across the entire frame. Very obvious if there are “gaps” in the blur because your brain will want to interpolate. If there are static/foreground objects, you probably won’t notice.
It's also fair game to critique these photos from an artistic perspective. Some are clearly intentionally staged and I argue that the messaging is a little clumsy. Sure, it's hard to avoid if you've filled your space with expensive design objects. Another comparison is cooking blogs where the photographers add visual clutter that looks good on instagram, but is impractical and unrelated to the food being cooked. The space itself is very nice, though you've got to be absolutely anal about keeping clutter down.
> A refused green card application might be the end of being ever in usa.
Do you have evidence for your other claim? The main thing you need to prove for a non immigrant visa or VWP is that you won't overstay or have intent to immigrate at the time of application and upon entry. Otherwise it's up to the consular officer like usual. You would need to declare the refusal/denial of course.
What will get you denied is "inadmissibility" if you don't submit a waiver. If you're inadmissible that usually means some serious violation and you've got other problems.
As far as I know, people have been successful in re-applying for EB green cards after being rejected when they've assembled a better packet.
Nothing new there, but under the new rules the former is no longer an option and you'd need to leave immediately. On the plus side consular processing tends to be cheaper and often faster (AOS and all the approvals vs the consular processing fee and a plane ticket).
That would depend how much "safety" is built into the control system.
The simplest solution I've seen is taping up the edges of the enclosure where you find gaps, to prevent heat escaping.
If it's only PID-ing the bed, the ambient temperature shouldn't matter. Less work to do for the bed heater. On the nozzle, it's similar. A 40 C increase in ambient temperature isn't much compared to the 150 C+ that the control system is maintaining. Since the active parts of the printer must be capable of running at the target chamber temperature, there should be no risk unless you exceed it. The question is really, is the printer designed to operate continuously with a chamber of X C?
However... the risk would be that if it's too well insulated there isn't a good way for the system to cool quickly if it needs to, or if it somehow messes with what the control system is tuned for. On the older printers you could re-calibrate the PID loops to your specific hardware and environment. The newer 32-bit firmware seems to not require user tuning at all. Similarly with full enclosures, you might worry about the power supply or other electronics which aren't meant to be run at high ambient (maybe fine though).
You could also look at a separate solution like enclosing the printer in well-insulated chamber, and aiming to keep that outer space above ambient. That would be a good option if you're expecting a big thermal gradient to your workspace, like an unheated garage in winter.
But lots of questions really. Do you want to run at a high chamber temp? Are you running in a cold environment and having problems? Trying to save power? These are different scenarios.
I think rotation is a bit extra though. The frustration with the dupont/0.1" headers is that they're not polarized. As soon as you add a key (or a literal magnetic polarity here), that problem goes away.
There are possibly ways to handle orientation in software. If you did a 4-pin and include VCC, the pinout on the pi is hi-lo-tx-rx, so you could look for hi-lo-hi-hi and wait for some period to identify any logic transitions on the tx pin. There should always be a period where that pattern is correct, because either the output is idle (high) or it's sending a bit sequence which will also involve a high somewhere. If you ever see the "2nd" pin high, you know it's not ground. Then your serial transceiver needs some way of setting up rx/tx or tx/rx (either you remap the pins, software serial, or use an analog multiplexer).
The Mk3 is also easy, and can be had for cheap now, but it doesn't have auto Z-adjust which is really nice. It's also noticeably slower compared to the latest models.
The downside is they're very expensive, relative to other earplugs and mine no longer seal as well as they used to so I'd need to get a new pair. They're still better than nothing. I started using earbuds around the same time, from using cans, and I wonder if I've very slightly widened my ear's opening.
I also use an eye mask if I'm somewhere that doesn't have good curtains or blinds. Really works very well, but I recommend one that wraps around and doesn't have an elastic band to dig into your ears (Matador makes a good one).
And yes, they're magnetically polarized according to the datasheet. To answer the sibling comment, they're through-hole soldered but frustratingly are 2.7mm not 2.54 so you'd be squeezing a little into an existing header (ignoring clearance on the side anyway). I'm not sure why the 7-pin is obsolete, they make up to 5-way active.
Or Adafruit? https://www.adafruit.com/product/5468