They could of course, but cellular modems with SIM cards are easily recognizable components which nobody appears to have seen on the drones. They also require a data plan with a national carrier which costs money. A satellite link is even more expensive.

Far easier to simply use the customer's cellphone, which already provides connectivity for free.

I'm not suggesting that a school district should homebrew their campus security system. I'm pointing out that the components are not expensive and they are getting ripped off.

"The surveillance system spots multiple threats per day, the district said."

I'm wondering how many of these 'threats' are legitimate and actionable, and not false alarms. But aside from that, surveillance cameras are already pervasive and you can expect to be visible on one almost everywhere you go in an urban or suburban setting. Video doorbells and home security systems have views of the street. Retail stores and any large building will have security cameras. Some big box stores I've visited have a mobile surveillance tower in the parking lot. It's just a fact of modern life. "32% of students surveyed said they felt like they were always being watched". No shit, sherlock.

And I can understand the motivation from the perspective of the schools, they are responsible for student safety. Even simple NVR systems (network video recorder) can at least document incidents of bullying behavior, instigating fights, locker theft, vandalism, etc. Automated counting of the number of people entering and leaving a school building would be pretty simple and useful. Automated recognition of exposed firearms seems very doable.

But I know for a fact that it can be done cheaply. You can get a decent network-connected surveillance camera for under $100. My home system has several that feed in to a $150 mini-PC running open source software. Yes there will be additional compute horsepower required for real-time AI analysis but annual district expenditures of "$4.8 million on security, including staff" seem amazingly high to me.

The drones by themselves will not be able to reach China on their own. Anything coming from them that could be sent elsewhere would have to go through a cellphone app that runs the drone, and that's where the risk is.

And then there's the possibility that there's a kill switch implemented in the app, but that's the case with any drone hooked to a cellphone.

>> LLMs are really good at automating the process of turning 1000 lines of otherwise approachable code into "enterprise grade" crap nobody wants to deal with

I can relate to that. Sometimes I'll be working on some hunk of existing legacy code that's easy to read but it is brittle and has always caused problems. I'll ask the LLM to suggest improvements, they are good at that. What typically happens is that every input and variable is now sanity-checked before use. There are additional try/catch blocks. Large functions are broken out into smaller chunks. Every possible failure mode is accounted for. The code is now painfully thorough, there's twice as much of it and it looks like it was written by someone with OCD, but it is rock solid.

>> Browse this thread to look for examples

I don't count on slashdot commentary for an accurate assessment of the state of the art. Some negatively biased curmudgeon who dabbled briefly with one of the low-grade models and had a bad experience is not representative. Contrary to what the peanut gallery here says, "76% of professional developers either use AI coding tools or plan to adopt them soon, with 62% already using them and 14% planning to adopt them" so it must be working out pretty well for most.
https://www.secondtalent.com/r...

Some truly amazing models have become available in recent months that are very capable indeed. Using them successfully is not rocket science, I can explain how in a few sentences so I don't think there's much money in it.

If it is a large new project or revision I typically start off describing what I want to accomplish in a couple of paragraphs and tell the AI to write a development proposal. I may not like what I see, I'll do back and forth refinements until I'm satisfied. Then I tell it to generate a phased implementation plan with small increments and testing at each step. I manually tweak that markdown as needed. Eventually I tell it to implement phase 1.1 of the plan and I walk it forward from there. If things go awry I can back off to an earlier phase and try a different approach. The AI can see exactly what is expected at each step, and this is how you can get 6k lines of high quality code in a day or so.

Admittedly this process takes an amount of patience that some people will not have, but if that's the case they are in the wrong business.

I generally do use Windsurf, but now also Google's Antigravity which is similar and their recent models are very good.

I totally get it that people here are suspicious and even hostile to AI. This is work that humans used to do and it is unsettling that machines can now do most of it.
But clearly things have changed and like it or not, a Stanford graduate who wants to keep up with developers like me and you is going to have to jump into AI coding assistance with both feet. The job interviews of the future are going to require you to show competence in causing a working application with at least 1,000 lines of code to happen in about 15 minutes.

I'm not going to post personally identifying information here, but I will share this screenshot from the dashboard of the AI system I frequently use. As you can see, I often get ~2k lines of code per day and sometimes as much as 6k. And mind you, this is production-quality code plus documentation that has been incorporated into a working system comprised of multiple distributed microservices.

According to Perplexity, "Modern practitioners and engineering managers often report that developers may commit around 50–100 lines per day on average".

https://imgur.com/3TC4jMK

I've got a lot of software development experience and I used to write a considerable amount of code by hand. Over the past few months that amount has dwindled considerably. Now I just prompt AI to do most of the coding. I compose the prompts, review the work, sharpen up the requirements, run the test cases, and guide the debug sessions. It takes some coddling at times but I frequently get at least a weeks worth of work out of it in a day with full documentation. 5x performance is not at all unusual if you know what you are doing. Like it or not, this is a significant paradigm change.

I don't really care much about how well you could write the code yourself. We don't need you to do that anymore. What I care about is how much valuable work you can get done coupled with a very large boost of AI assistance. Someone straight out of Stanford had better be 'AI-savvy' at this point in time as part of the package. If you are a smart, well-educated person who can make it sing you will be well worth hiring. Otherwise no.

Take away all the subsidies and we'll see how much production the market will support. Reportedly about $30 billion annually.

https://www.cato.org/briefing-...

"Most direct subsidies are for large producers of corn, soybeans, wheat, cotton, and rice"

>> If you look at it the right way around

Who said you get to decide which is the right way? Your cite clearly states;

"the body of scientific evidence indicates that increased extreme weather resulting from climate change will instead reduce crop yields, making food more expensive"

>> the agw global warming advocates have INSISTED that food crops wouldn't flourish in higher CO2 environs

The Yale report explains it clearly, what didn't you understand? The effects of extreme weather outweigh the benefits of higher CO2. Droughts, floods, heatwaves, etc.

'A 2020 analysis of 30 years of FACE experiments found that higher carbon dioxide concentrations increased crop yields as expected when “under non-stress conditions.” But when stressed by factors like changing temperatures or precipitation, as climate experts said recently, “the yield increases were suppressed and in some cases erased.”

According to their website, they "developed a brand-new optical system that uses ultrasmooth, multilayer mirrors inside a vacuum chamber. Each mirror has over 100 layers of materials that were carefully chosen and precisely engineered to maximize the reflection of EUV light.

Flatness is also crucial for EUV reflection. The mirrors are polished to a smoothness of less than one atom’s thickness. "

"The prototype, completed in early 2025 by former ASML engineers who reverse-engineered the Dutch company's machines, is operational and generating EUV light". "recruits are working under false identities inside secure facilities"

China must be paying them a fortune for this theft, no wonder they want to remain anonymous. The basic technique ASML uses to create the EUV light is public knowledge and ASML even describes it, but the machinery to do it is not.

They shine a powerful industrial CO laser at microscopic droplets of molten tin inside a vacuum chamber. When the laser energy is absorbed, the tin is turned into an extremely hot plasma that emits extreme ultraviolet light with a peak wavelength of about 13.5 nm.
https://www.asml.com/en/techno...

I assume the thieves have been able to replicate that part, but there's a lot more to it. Apparently most of the light is lost. They can't use any refractive lenses so it is all channeled by coated mirrors.

"To get enough brightness for high-volume manufacturing, the tin-droplet and laser interaction is repeated up to about 50,000 times per second. The combination of high repetition rate, precise droplet positioning, and efficient collection optics produces the EUV power at the scanner input needed to expose hundreds of wafers per hour."

Someone or group will need to be able to assign access I agree. But in this case "perpetrator had once worked in South Korea as a software developer for authentication systems" and "the suspect secretly held on to an internal authentication key that granted him unfettered access to the personal information of Coupang users". There should be no such key, and obviously they should be able to revoke all access keys a previous employee has held.