1) Sprout the streetlamp from the box, add a car charging port.
2) Hire an artist to paint the box with artwork (property owner gets to pick from option set). Consider it to be city owned art installations.
453 karma · joined August 29, 2019
glwittel [at] gmail [dot] com
1) Sprout the streetlamp from the box, add a car charging port.
2) Hire an artist to paint the box with artwork (property owner gets to pick from option set). Consider it to be city owned art installations.
I really wish we'd just have municipal ISPs here. I'm only a few miles from Google HQ, but the internet options are crap.
I took it well over a decade ago, but the current material is largely similar. Going in you were required to know C and C++, and math pre-reqs (Calculus, linear algebra). Other university classes might skip the rasterizer building, but I think its a useful foundation.
We started with basic OpenGL to render models. Then we replaced the OpenGL with our own rasterizer, adding features as you went. It was a tough class (10 week schedule), but very rewarding. Probably the worst part was debugging rendering issues. You could see the problem, but it was hard to locate where your math went wrong.
On top of that you need explainability and reproducibility. Depending on the ML in use, this can be difficult. When a FP or FN occurs, customers want to know what happened and can you fix it. They don't like to be told things like "because the ML said so." Reproducibility is important for support. When things go wrong, they need to know how to right it. If you can't explain or reproduce, its hard to say the misclassification is resolved and won't happen again.
Ignoring the burden shifting of on call rotations, I often find the young mavericks don't sleep at night -- they're fire fighting and debugging their work.
With the plethora of options, what missing for me is what ones perform well under heavy loads. Its painful to find out after the fact since each ingress controller deploys in differing (and sometimes not compatible) ways.
For example, with nginx-ingress there are gotchas under heavy load. nginx-ingress doesn't support SSL session caching on the upstream (nginx<->your pod). This is a deficiency in the lua-balancer implementation. You can tune keep-alive requests on the upstream, but it isn't always enough. That 50% CPU savings from SSL resumption is costly to lose at times.
This has bitten me when a client side connection burst requires a connection burst in nginx<->service. Upstream services then burn a lot of CPU negotiating SSL, to the detriment of request processing. This then causes more nginx connections to open up due to slower request processing; and might cause healthchecks to fail. There just aren't enough tuning parameters to control how hard nginx hits your upstream PODs.
How much do 3rd party/FBA sellers participate in prime days’ official deals? I could see manufacturers doing their own deals, not sure about others.
The first few prime days had a lot of real deals on name brand stuff. More and more, deals on name brand items are less common or supply constrained. Other 'deals' were fake deals (discount off artificially high list price) or deals on random no-name products (or garbage products).
Typically you synced the equivalent of headers and bodies separately (for the groups you were interested in). This way you only grabbed what you wanted. For binary groups, people settled on patterns in subjects, etc so that clients could automate downloading of files (binaries had to be encoded and split into parts due to post length limits).
On regular text type newsgroups you might set your client to download headers+bodies in one shot for offline mode. With offline mode you would then just download what's new vs your cache. Then work against your local cache for reading/replying. Once online again, your batched replies could be sent.
If your ISP didn't have the newsgroup you wanted, you would have to point to another NTTP server. For a while there were public ones that anyone could use. Over time, those public ones went away or dropped problematic areas like alt.binaries.*. These days Usenet is available as a for-pay service.
The less glib answer is that while best practices have gotten better, companies aren't necessarily better at following them (your company or your vendors). Additionally, on average people have not gotten all that smarter (while scammers have become more sophisticated).
Its all about defense in depth. Plan for compromise, trade off the costs (nothing is free).
At a technical level there are basic things like: separating dev from production, requiring 2FA across the board, strong passwords, certs, separation of privileges, up to date software, secret management, etc that form strong defenses.
Another way to approach this is to pose scenarios and understand defenses:
* What information could an employee give away that would then lead to a compromise? How do you stop it? Password to a phisher? Check in credentials to github? Wire money to a fake vendor/wrong account?
* If X was compromised, what is the impact? What would it take to recover? Where X is: dev box, production box, customer facing service, etc.
Two insidious scenarios that happen way too often:
* Latest malware distributed via legitimate sites (ad networks, sharepoint, dropbox, etc) -- Hard to prevent beyond up to date systems and content filtering. If paranoid, all browsing through VMs or other forms isolated environments. It won't stop compromise, but it will restrict scope.
* BEC -- Attacker studies your company. They impersonate an employee or partner (down to linkedin profiles) and convince someone to wire money. Or worse -- account compromise of a trusted partner. The attacker then uses this partners' account to hijack an email thread or send a 'normal' request like: "Our account has changed, future payments goto account YYYY."
Which brings us to vendors. Can a vendors' weak security break your company? Is it a good idea that your Microsoft Office 365 account allows unlimited password attempts (at least MS used to by default)?
Things changed when my dad bought a modem in 93 or so. We had Compuserve but it was expensive. Through Boardwatch magazine I started to try out local BBSes (916). Eventually a few like 24th Street Exchange and this BBS sucks became my online homes. 24th was particularly interesting as it was large (10s of lines), had Fidonet and possible Usenet feeds. Oh and it ran on a BSD. I learned a lot there.
Two key inflection points came:
1) I saw my first demo which made me want to learn to code in 94 (heartquake by iguana ; thanks guys!)
2) around 94-95 I heard about Linux and ended up with a Slackware disc from Walnut Creek.
In 1995 came full internet access with a whole new set of rabbit holes. Usenet, cipherpunks, IRC, etc.
Over times my interest varied as a teenager would. Upon starting college in 98 I had amassed a lot of experience - coding, graphics, graphic design (2d and 3D), web dev, Linux, building computers, etc. At the time I didn’t realize it, but this constant exploration and learning helped me build a huge set of tools that have served me well in my career.
In the end, the biggest lesson? Learn. Follow your curiosity. Roadblocks? find a way around, worst case, spot the dead ends and try again.
> "I'm hoping they learn different words within colonies - but if they teach the others bad language and I end up with 250 swearing birds, I don't know what we'll do," he added.
This will be an interesting experiment.
Largely we had to choose -- take the college prep classes (AP whatever, languages) OR trades classes. There was definitely classism -- "trade classes are for poor kids". =(
I would much rather schools require things like: financial literacy, and building/repairs/cooking. They are lifelong skills and very useful. There's no reason its an either or proposition.
> ... placing the Far-UVC lamp 24 centimeters above the surface of the plates.
Proper UV sterilization takes time (minutes, hours). I'd worry that many people will just wave a light and think "its clean". It doesn't work that way.
In our case most of the work is done on paper and they hold it up to the camera for the teacher to see or take a picture and submit it.
Our first grader is still adapting. Historically he got maybe 20m of screen time a day max. Now it’s about 4.5 hours. There’s a lot of movement breaks and some non screen activity (quiet draw or read), but it’s still a lot of screen time. He’s often pretty bored.
For preschool when they could not open for in person the most they could offer was about 30m a day. That was basically story time and sharing so kids could say hi. Many families opted out.
While the district mandated google classroom for all kids, teachers of younger ages have been given the option of using Seesaw. For young kids its miles better. Even as an adult I find Google classroom to be awful. It feels like enterprise shovelware with bad UX, inefficient workflows, etc.
A lot of it is both how much work the teacher puts in, and how much leeway the school/district give. At this age, it requires an adult nearby at all times. It may be to help him stay on task, occasional assistance (finding something, tech issues, etc.), and even just company of another person. We've had to juggle help from family, and my wife is not working just to keep things going.
We were lucky last year to have an amazing Kinder teacher who did a lot to try and hold things together for the classroom (including 1:1s, summer penpals, and a lot of things on her own time). This year is a WIP, but our school has been very receptive and flexible. I worry most about the kids with two parents who must be at work and cannot give the same level of support we've been able to. They are likely to fall behind :( Thankfully teachers are trying to help those kids the best they can. It is not easy.
Last spring while my son was finishing kindergarten it was extremely difficult. Seesaw helped a lot because they enabled community sharing so kids could see each others work and reply to it. We had to put a lot of effort in to keep him engaged, doing as much work as possible, etc. There were many meltdowns over seemingly simple things like "write 1 word".
Preschool is different and in a tougher position. The primary goal is socialization and kinder-prep. Both are largely out the window for online only. Kids of this age just can't sit for that long in front of a Zoom type session.
In person is limited but now available. The main challenge is preschools are struggling. The regulatory burdens force class sizes to be a fraction of normal. Naps are basically non starter (physical space requirements), food is tricky. That means the pre-schools are losing huge amounts of money every month or laying off staff. To top it off, the recent fires create such pollution you have to choose: (1) Close windows and violate covid regulations (2) Close for the day.
The main challenge I have is not all engineers are equally willing/comfortable/able to communicate their ideas in this way. Sometimes because of ingrained preference (comments are never necessary), other times because of language barriers.
The downside of the SQLite dB is no network editing. I’d love to be able to edit from multiple machines without a cloud subscription or super slow NFS.
Moving to Capture One or something else seems worthwhile. Tried Darktable but wasn’t a big fan.
Worrying that the CloudSQL internals (like the private IP range) aren't strongly walled off. It will be interesting to see how this changes in response to the researchers' work.
Flavor wise it was not what I’d want to use for regular consumption, but if it works better when sick? Sure.
In the end I containerized the whole thing with defaults, logos, etc so we’d have standard docs with little effort.