Given that school budgets are absolutely gutted with mass layoffs this year and next, and the miscalculation looks like 2/3 of the budget shortfall, hiding such a basic and impactful error requires a much better explanation than I see in that article. It looks like it was done to stifle debate about budget allocations, which would be necessary in the circumstances.
The education system seems one of the only places where vastly improving technology over the past 30 years has not translated to cost savings or improved outcomes.
True, a kid who has had too much screen time is not good, just like a kid who has had too much lunch is not good. That doesn't make lunch bad, it just means the kid needs the right amount.
Screens can be helpful for kids (mine have learned a ton from Khan Academy and other online tools), but kids will have different thresholds. Some will only be able to learn a little from screens because they can't work independently. Others can learn a lot. Blanket statements like "kids lean better from humans than machines" are not helpful. They obscure the fact that there is typically one teacher for 25 kids, whereas there might be 25 screens. Even if a screen is only 1/10 as good as a teacher, it could be that learning from a screen is better than learning from a teacher (who is busy with your classmates almost all of the time).
My kid learned more math when she was doing AoPS for 2 yrs than when she was in class listening to lectures she already knew, followed by worksheets she had already mastered. Machines enable much more differentiation.
Your anecdata has a gigantic blind spot of you being a capable and engaged parent that monitors and guides your children’s screen time to be educationally productive. That time is a luxury, and your expertise is rare.
To take it back to the lunch analogy that you provided, it’s a bit like saying, “I don’t know why everyone’s kids are hungry at school, I pack a nutritious and filling lunch that I know my kids enjoy every morning.”
Logically speaking, this is incorrect. OP said that humans teach kids better than machines, full stop. To disprove this claim, I do not need to show that all kids learn better from machines, just that some do. I have shown this, and even admitted that it doesn't work for all kids.
Also, I didn't need to monitor/guide my kids so they could learn from AoPS or Khan Academy. Those platforms are self-guided. But regardless, my kid learned pretty much zero from school math, so the threshold for "better than the teacher" was very very low.
I’m going to sidestep your pedantry and focus on why we’re actually having this conversation: it is obvious from data and observations at scale that children struggle to learn meaningfully in the presence of technology, i.e. screens, in the same way that children learned with educators of the quality that taught previous generations (higher salaries).
Techno-Utopianism is such a grating ideology, especially when thrown into the ring alongside all of the other garbage that education experts have to deal with when trying to enact meaningful change in the education system.
In my country every few years a bunch of people invent some new method to teach kids how to read.
Completely ignoring that humans have been doing exactly that for at least 5000 years.
Gains in efficiency due to technology improvement, as far as I know, are always attributable to fewer people being able to do more / produce more. There are some things that will never have gains in efficiency. Technology will not reduce the number of players on a football field from 11 to 10.
My 1,500-student public California high school currently lists 7 administration-team members (principal, executive assistant, three assistant principals, school-facilities manager and food-services manager) and 11 administrative-support members (school data-processing specialist, print-center technician, senior-clerical assistant, separate registrar and attendance roles, interventions-support specialist, and others). That doesn't include 4 site maintenance, a network-support and a separate network-systems specialists; a separate media-library specialist; 2 psychologists; a college and career advisor; 4 school counselors; a wellness-space support specialist; and a social science and an athletic director.
34 administrative hires. One per 44 students. Many of those roles strike me as fluff.
It’s because there are tons of laws and regulations regarding minors in school, and administration tends to be homegrown (initial expertise in teaching) rather than explicitly developed to navigate the social, political and legal landscape. I’d wager that more than half of those positions are “best practice” staffing decisions in response to this landscape. A handful might also be due to expressed needs and wants of parents. Likely wasteful overall, but students, teachers and families would likely feel the impact and not be satisfied if any positions were axed.
> Which of those roles specifically would you say are fluff?
Food-services manager (it's all oursourced to Aramark), data-processing specialist, print-center technician, senior clerical assistant, one of registrar or attendance, two of site maintenance, one of the network specialists (probably both–one across the district is plenty), and probably at least one of the counselors and the separate social science & athletics person, who should just be one of the physical education teachers. That's about ten people, or a million dollars–minimum–in annual savings.
As a full-time position? Aramark literally ran the lunch counters. I could see it being a district-level position, though it would be better positioned as a general procurement role.
> Data-processing specialist and print-center technician both sound like fancy names for secretarial roles
I agree. I was suspicious when I didn't see a secretary for each of the assistant principals listed.
> You're honestly saying schools need fewer counselors in what has been generally regarded as the worst generation for child mental health in years?
I am. Unless the counselors are constantly doing actual therapy I'm deeply sceptical you need that many for a student body of that size. The fact that they're assigned based on the first letter of your last name versus anything remotely thematic or behaviour based seems to emphasise that hypothesis, for me at least.
(When I went to the school, there were bullshit jobs everywhere. One of the counselors didn't deign to meet with students. Her role was "strategic" or some nonsense.)
> One of the counselors didn't deign to meet with students.
Why would she? That'd distract her from the actually important work of fabricating the reports that make her looking amazingly competent.
My mom is a retired teacher and her main complaint during the last 10-15 years of work was that with all the bullshit paperwork they're required to fill, the teachers literally don't have the time to just plain interact with the students. You want to make an odd, unscheduled extracurricular event? Waste a small pile of paper before organizing it, arguing for how amazing it will be for the students' education, and the an even larger pile of paper afterwards, bluntly telling just how amazing it all worked out and checked some tick boxes the upper management cares about.
This is not creating widgets or lines of code, not creating a product for consumption, this is fostering the development of inquisitive minds, hopefully encouraging them to become critical thinkers and ultimately the next generation of leaders who will push the bounds of human knowledge further than ever before.
Why would better tools be expected to do enable teachers to do that for more students at a time?
There is a lot of research out there showing worse educational outcomes as class sizes increase. This is one of the areas where wealth disparities in education manifest; rich areas tend to have smaller class sizes, and historically the very rich would pay for private tutors for their kids, whereas poor kids are stuck with bigger class sizes, less individual attention from educators, and typically average worse educational outcomes.
>This is not creating widgets or lines of code, not creating a product for consumption, this is fostering the development of inquisitive minds, hopefully encouraging them to become critical thinkers and ultimately the next generation of leaders who will push the bounds of human knowledge further than ever before.
There's plenty of drudge work teachers do that's not "fostering the development of inquisitive minds". Grading papers, preparing lesson plans, etc. I don't see why not at least some of that can be offloaded to AI.
> Why would better tools be expected to do enable teachers to do that for more students at a time?
Khan Academy showed that one great teacher distributed to millions does that pretty well. It doesn't make sense for every teacher in the country (the worst and the best) to create their own syllabus and teach the same thing over and over again.
Sure, let's have 100 child classes which are hell on earth for everybody involved, starting with the little kids who will literally be scarred for life from it.
Teacher costs should be going up as much as we can afford, to keep reducing class sizes as a fundamental part of quality education.
I agree that admin is ripe for efficiency gains. A local school district cut dozens of teaching roles, not even one person from their extremely bloated central administration. It's also out of touch with the schools with no campus visits, and serves mainly as a hindrance to any sort of actual work going on inside the individual schools. It's a horrible caricature of bureaucracy.
> Only if you assume if per-teacher productivity can't increase.
It can't.
The only axis upon which teacher "productivity" could increase is by increasing the size of their classes.
Every study and every practical example of doing that ever done shows that it negatively impacts student outcomes.
Not because the teacher is failing to be whatever it is you imagine "more productive" to be but because there is a minimum amount of attention needed per student for them to not fall through the cracks and one person's attention is not scalable.
> only axis upon which teacher "productivity" could increase is by increasing the size of their classes
And hours in class. Or productivity of time in class. I'm not saying the former is desirable or latter feasible. But the education "production function" has three inputs.
Oh, it definitely can, in a way very similar to the way you can dramatically increase doctor's success rates by being selective about who you treat.
Specifically: take the most disruptive students and eat them. (Be stealthy about it, the point is not fear of punishment.) The productivity difference between a classroom that spends 90% of its time on instruction vs 90% of its time on classroom management is massive.
That's why you have to be careful about applying business notions like "productivity" to governmental duties like education and mail and highways. (I dearly wanted to include healthcare or at least hospitals in the list, but I live in the US.) Businesses can and should be selective and take higher risks. For governmental tasks, productivity isn't even well-defined. If you're failing (or eating) 20% of your students but the other 80% are doing amazingly well, is that better or worse than 99% of everyone doing just okay? How about if everyone's test scores go up and practical ability goes to shit? (This is not a hypothetical, not where the kids have figured out how to use ChatGPT even for the tests. Which is a lot of places.)
Teaching is nowhere near Pareto optimal right now, so I'm not arguing in favor of the status quo. I'm just saying you have to be very, very careful when pushing for "productivity".
You could increase per teacher productivity by running 12 months of school per year, but you would increase costs; and in some parts of California, you would need to rebuild schools with air conditioning to hold classes effectively in the summer.
Covid showed distance learning doesn't work for most kids. So you can't eliminate real estate costs or hire educators in low cost areas. Computerized education doesn't seem to work, either.
>You could increase per teacher productivity by running 12 months of school per year
Productivity is output divided by some input, either labor or money. Working for longer isn't going to magically increase productivity.
>Covid showed distance learning doesn't work for most kids. So you can't eliminate real estate costs or hire educators in low cost areas. Computerized education doesn't seem to work, either.
Right, I don't have a specific solution for increasing teacher productivity, but it's not obvious that it's a law of economics that it can't increase. People thought lawyers and doctors couldn't be automated away, then came chatgpt.
> People thought lawyers and doctors couldn't be automated away, then came chatgpt.
Form contracts and will generators and what not was automation for lawyers. Plenty of enter symptoms to get a diagnosis stuff out there for doctors; or the more paletable, enter symptoms for charting, get a suggestion and enter medicines and get alerts about interactions.
in California there are not more students.. all tiers of schools show falling enrollments, year after year. Except community colleges, where they have discovered that more than 15% of all students are ghost enrolled.
--
California K-12 public school enrollment fell by 74,961 students (a 1.3% decline) for the 2025-26 school year, marking the largest drop since the pandemic. This loss was significantly higher than the state’s Department of Finance projection of only 10,000 students.
The decline is driven by lower birth rates and a reduction in immigration, with the latter exacerbated by families fearing enforcement actions. Los Angeles County accounted for nearly half of the state's total loss, losing 32,953 students, largely due to a decrease in newcomer students within the LAUSD.
Private schools saw a steeper drop of 6.6%, while homeschooling declined by 3.7%.
The enrollment drop is causing budget deficits, leading to staff layoffs, program cuts, and potential school closures.
Hispanic students experienced the largest numeric loss (48,064), while white students saw the largest percentage decline (2.68%).
English learner enrollment fell by 8.2%, partly due to reclassification and partly due to out-migration.
From glancing at the numbers it looks pretty similar, but there's been a huge drop in births in California in the last 10-20 years so it's probably the last few years where that will be true.
Peak birth year was 1990 after booming through the 80s, births started falling off a cliff after 2008 and last year there were about the same number of births as in 1980 despite the population increasing by 80%.
It shouldn't be surprising when you see what the educational technology money is actually spent on. Ive seen like a dozen "smart boards" purchased by my schools from the 00's and never once seen one used. Or how about the completely dog shit computer testing software that fails one-two word answers based on random capitalization or punctuation with no indication on what they expect. Each thing sold as the solution to all the past problems, but introducing more problems itself without actually solving any old ones.
I already posted this, but the budget cuts are also because Federal funding was up to historic level due to COVID and has been cut significantly below baseline by the current administration. California should not calculate school funding the way it does anyways. A teacher is paid the same regardless of whether their class size is 20 or 30, whether some kids are home sick or not. So a more complex allocation system that takes this into account is absolutely required. Failure to do so is intentionally under-funding schools.
My point was simply that it is shady to hide the miscalculation, and I indicated the point in the budget under severe stress right now I would have liked to see funded should the appropriate discourse have been possible. The cause of the stress is irrelevant. 2 billion goes a long way when everything is being cut. It could save 2 billion in services.
It is just a fact that California schools are laying off a large percentage of personnel and getting rid of many programs. Pink slips by the thousands have been sent out that will take effect in a couple months at the end of the school year. If you don't know that, you are not informed.
Those links are completely irrelevant because they are out of date. Budget had temporarily increased due to the availability of COVID funds, and now there is a very harsh snap in the other direction. Shortfalls are directly linked to actions by the Trump administration, and their downstream impacts. Every state needs to step up and deal with it.
A quick google search of the UNSECO target is "at least 15% of total public expenditure (or 4–6% of GDP)" and both the US (~5%) and California (~4-5% of gdp) already pass that criteria.
As far as I can tell (including looking at third party analytics attempts), there had been a massive increase in users over the last 3 years. Smaller communities tend to hold their trademark character a lot better. Pure speculation, but (beyond the bots) I suspect that a lot of the newer users are younger, and the attempt to be a bit more focused and sincere here is something they miss before they start posting.
> what I call "red" could be very different to someone else's subjective perception
It's worth noting that is true of virtually everything we know. >>This is a very simple sentence.<< Anybody who understands English, 'understands' it. But what it means to understand it is perhaps completely different for each person. As long as they fit into the same place in their worldview (Lewis Caroll's Carrollian syllogisms come to mind), practically it often doesn't matter beyond recognizing the wonderful uniqueness of each human being. Likewise, unless somebody is color blind or perceives more colors than others (tetrachromats), it doesn't matter since the relationships between the different concepts or colors will be analogous amongst most people - so a common understanding within the differences is possible. Or perhaps it is more precise to say that there are so many data points in color perception or anything we know, that despite the minor differences in relationships, we understand each other because the differences must be minimal given the practically unlimited data points constraining our perceptions. In fact, when people's perceptions of things vary too much, they can be classified as mentally ill even if they understand many things perfectly well.
At the same time, there's some commonality for what words mean in different contexts. For example, even though we all have our own experiences with the concept of "dog", there's a common core where we have enough of an understanding what other people refer to as "dog" to allow discussing the concept. Likewise, for most people, dog is more similar to cat than to house.
Imagine if we could build a machine that reads a bunch of texts and tries to extract this meaning by looking at which words commonly co-occyr with other words in different contexts. Perhaps something interesting would happen...
Yes, but the qualia could be completely different and we'd never know.
For all I know you don't just have a completely different experience of red, but a complete different experience of geometry and spacetime.
Your subjective experience of vision could be a mirror of my own. But we'd both still associate "right" with the same half of the body.
You might not "feel" curves and lines the same way.
As long as everyone's mappings and weights are identical, the qualia themselves could be anything.
We assume the qualia would at least be recognisable, and they can't be too different because there has to be a common core of experience categories, with recognisably consistent relationships.
But beyond that - anything works.
This isn't a hypothetical because once you get into politics and ethics, the consistent relationships disappear. There are huge differences between individuals, and this causes a lot of problems.
I am so embarrassed I have never tried it. I am extremely bursty with video so I just grabbed obs and openshot and use those. I always presumed it wouldn't be enough because it was 'just part of the kde suite'. I will try to remember to spin it up next project.
I have a very newcomer-type question. What is the output format of your plan such that you can break context and get the other LLM to produce satisfactory results? What level of details is in the plan, bullet points, pseudo-code, or somewhere in the middle?
I just tried littlesnitch and it did not resolve very many ips to domains, which is pretty basic. It also failed to identify most processes, and they were grouped under "Not Identified". It appears these are known limitations of the Linux version [1]. So for that alone I need to stick with opensnitch.
[1] "Little Snitch for Linux is built for privacy, not security, and that distinction matters. The macOS version can make stronger guarantees because it can have more complexity. On Linux, the foundation is eBPF, which is powerful but bounded: it has strict limits on storage size and program complexity. Under heavy traffic, cache tables can overflow, which makes it impossible to reliably tie every network packet to a process or a DNS name. And reconstructing which hostname was originally looked up for a given IP address requires heuristics rather than certainty. The macOS version uses deep packet inspection to do this more reliably. That's not an option here." -- from https://obdev.at/products/littlesnitch-linux/index.html
Regarding unidentified processes: Little Snitch daemon must have been running when the process started in order to identify it reliably. It's best to reboot after installation so that Little Snitch starts before everything else. I should probably note this somewhere.
And regarding failed reverse DNS names: Little Snitch is sniffing DNS lookups. If lookups are encrypted, there is little it can do. We usually recommend DNS encryption at the systemd layer, not at app layer. This way we can see lookups on 127.0.0.53 and the actual lookup sent out is still encrypted.
Also, it's currently only sniffing UDP lookups, not TCP. The eBPF part is already very close to the complexity limits (700k instructions of allowed 1M) and adding TCP parsing would exceed this limit. It should be possible to forbid TCP port 53 with a rule, though. Some complex DNS lookups will fail, but routine things should still work.
Not all "hostname lookups" by applications happen over DNS (or the DNS is done by something like systemd-resolved, which is often using encrypted lookups), so in many cases, depending on NSS configuration (e.g. 'file', 'resolve', 'db', 'nis', 'mymachines', 'libvirt', 'winbind', ...) this would never work?
Yes. For these cases it won't work.
OpenSnitch intercepts the client side library for this reason. I would rather want to avoid this for the moment and wait for feedback.
The thing is, 127.0.0.53 is a fallback. The real default upstream is nss_resolve, which talks to systemd-resolved via non-DNS protocol on a UNIX-domain socket. Ubuntu disabled this in favor of the less-featured fallback. If you insist on sniffing DNS, you need to add instructions to disable the native nss_resolve module by not including it in /etc/nsswitch.conf.
Thanks for that hint! We still get the lookup if it leaves the machine unencrypted, but if you have both, the Unix domain socket and DNS encryption, we miss lookups.
If I don't know who my machine is talking to, the information is not very useful. So there needs to be a fallback on some level.
Perhaps there should be a mode where littlesnitch just does its own lookup using the system-configured rDNS, for example from the ui or for specific processes, etc? It should be cached if it is a recent lookup, so minimal performance implications; and offloaded to the system rDNS resolver, so minimal instruction set.
We do not want the reverse lookup name. For instance, if you look up a google.com name with dig, you get an IP address. If you then do the reverse lookup with dig -x, you get a 1e100.net name. That's as good as the IP address for our purpose.
Plus: We need to respond with a DROP or ALLOW verdict to a network packet without the ability to do any blocking requests. So we can only use information already available in the kernel to decide.
I guess that makes sense, since it's pretty new. OpenSnitch is great software in terms of functionality but I find the UI lacking. If LittleSnitch can keep the same functionality, while improving the UI, I'm switching. My other current concern here is that the LittleSnitch UI is just a Webview and I think it would be much better if there was a native option (ideally GTK-based for me, but Qt would also be acceptable). Webviews are slow and full of bloat.
Every time I try to change my user agent with a FF extension I get hit with brutal cloudflare captcha loops. How are you changing your user agent in a way that this is not a problem?
I had a problem to fix and one not only mentioned these "logs", but went on about things like "config", "tests", and a bunch of other unimportant nonsense words. It even went on to point me towards the "manual". Totally robotic monstrosity.
Unbelievable. Some dude makes an hn account after lurking who knows how long, makes his very first innocuous comment a day later, and is immediately attacked as a newb or a shill? Give people at least a little benefit of the doubt.
reply