j4k0bfr

I'd assume that most datacenters use evaporative cooling, since it does consume water directly and is pretty cheap. Some web searches on the topic were swamped with companies trying to pitch their solutions (as if I'm in that market??) but this[1] link looks... slightly less suspicious.

Dealing with waste hot-water sounds like a PITA. Piping it straight back to the reservoir would need a dedicated pipeline, probably with insulation so you don't bake everything around it. Harnessing the excess heat on-site would be awesome, although I'm sure datacenters would already doing this as best they can to recoup electricity costs.

[1] https://www.datacenterknowledge.com/cooling/evaporative-cool...

I wouldn't be surprised if many of the motorsport software engineers you mention come from mechatronic/electronic/aero backgrounds. Don't get me wrong, I've seen an analog electronic engineer perform software wonders; 10,000+ lines of bash, custom C macro libraries, and a PHP runtime older than I am have taught me that programming can be closer to art than science. But relying on that cursed bundle of chaos felt deeply violating and unsafe lol.

Nowadays, I mentally place any engineer without production software experience into a 'research' box. Any resulting code may break the laws of physics, and should be tested as such ;).

Edit: It's also a minor crime to look down on automotive software engineering! The stakes of mass-manufacturing have forged some seriously awesome software. Top-of-mind for me is CAN support in Linux and safety-critical contributions going into ROS2.

With something safety-critical like an ECU, I'd be surprised if they left formal proof out of their software design and testing. This would make unexpected hardware behaviour much more likely imo. I guess I'd get around it by setting up automated hardware-in-the-loop simulators for testing every possible environment and input. Surely the cost would be worth it for F1 as a whole.

At a startup I worked at, we jerry-rigged Pi 5's with commercial of-the-shelf environmental sim software to achieve this. Very cost effective, lol.

I agree with your first paragraph, a majority of corporate programming jobs probably suck. However I'm not convinced that AI will reduce demand for these positions to in a major way. It will instead change the required workflow. I've got a couple of industrial revolution points I'd like to share; The Jevons Paradox addresses demand concerns. Software is now theoretically faster and cheaper. More projects are now financially viable, since total unit costs are down! The other side is the Automation Paradox, where now even more specialised knowledge is required to fix and maintain AI systems.

Your disenfranchisement is valid, however fortunately (or unfortunately?) do-nothing soulless corporate programming jobs have a good chance of sticking around, if in a changed way.

Comparing programmers to farriers is definitely strange on the surface. However I think it's quite unfair to reject the authors premise due to the ephemeral nature of some software. And I think it's downright arrogant to assume moving from horses to cars wasn't seriously challenging for his ancestor. But yeah, I would love to chat with the old man about what that transition looked like.

Also, AI adoption might look like it's moving fast for individuals, however I see many companies struggling to actually utilise it effectively. Business processes are slow moving political machines. Sure a large part of the software industry is ephemeral, indirectly providing capital value. However many jobs are still very tangible; embedded, backend web systems, networking etc etc. Instability in these systems absolutely results in tangible loss of value.

It will absolutely take decades before we figure out 'correct' AI usage at an industry scale. Heck, even frontier labs are still undecided about best practices.

P.S. ty for the article link, will read!

What, it's a good analogy for this audience. I know enough to have an opinion though. You'll never see me using a thread to locate a part. And I know taste isn't the only difference between red and blue loctite!

If you think my reply was off-base, give me a bit more than this mate.

I agree that any military on equal standing would blast the SR71++ to smithereens. But how many wars does the US fight against equals? And on the reconnaissance side, there are pros and cons to satellites. They're hard to replace and easy to track & destroy. Plus even the best space sensors lose quality due to distance.

Do I think an SR72 is a good idea? Hell no, but I don't think like the US military haha.

Agreed, I can't imagine much scientific/engineering value that NASA would get from restarting the SR-71A by itself. There's an easier way to simulate the Blackbird's flight characteristics without having to magic the damn thing back into the sky.

I reckon it's much more likely to be used as part of a training program (maybe related to the shiny new F-47?) or for destructive/non-flight testing and research.

Not a mechanical engineer here, but I reckon close to a decade would be pretty realistic for full operational restoration. Even a small amount of lost critical maintenance knowledge could cost the whole airframe.

I'd compare it to a really old codebase. Sure, our modern tech is much better. But if you need to make an old, top-secret COBOL program work as-is, you're in a pickle. Sure, we could rewrite it in Rust and make it way better, but management sees it as more cost-effective to just "turn the old one back on, it worked great!". And unlike software, it's pretty hard to comprehensively test a plane in a non-destructive way. Bolt falls off turbine while in flight, etc etc. Losing even a single SR-71A airframe would be a massive headache, methinks.

This is a bit surprising to me, considering how much AI companies love to hoard data. Especially since some of these ad companies are direct competitors!

My best guess is that these ad mechanisms are a bit rushed and/or that investor demands for profitability are fighting against company self-interest.

Edit: I guess some data will always need to be leaked for AI chat ads to be most effective. But I imagine AI companies would rather deliver the targeted ads themselves rather than letting competitors do it for them. It would be scary to see AI companies become ad companies too (instead of just hosting them).

Please educate me on what other options are viable for keeping kids off social media.

This "ban all kids" kind of take is bananas to me. It implies you either A) believe we should trust famously secure and benevolent tech companies with government ID or B) believe parents have the time or technical know-how to exhaustively monitor the internet access of kids. Even as a software engineer, I've found it insanely tiresome to try operating half-assed parental controls across devices.

Mostly vast amounts of sexism, but I also see racism, antisemitism and islamophobia waaay too much. The 'isms' tend to glom onto certain posts and go nuts. Algorithmic feeds at their finest, I guess.

You can enable viewing dead comments in your settings.

It's a bit frustrating to see this video immediately sniped while Meta continues to host some of the most insidious content on the internet.

It actually scares me that kids can read IG comment sections.

Agreed! Although I like the idea of considering each logical branch added to my programs. It kind of forces me to think through how variance is being treated in my call stack.

I've never done workload optimisation myself, but have looked into it just-in-case. It's a really interesting field and unfortunately, you need to know about both your compiler and target CPU/GPU to get the big gains. I believe most modern CPUs have registers for cache use. You have to do some guesswork to get IPC numbers, since core frequency can vary.

Increasing instructions-per-clock is all about minimising program branching (essentially 'if' statements). Because a CPU core can execute instructions faster than main memory can fetch em. It's a fun game to look at an 'if' statement and figure out how you could instead make it an arithmetic operation :).

Maximising cache hits is all about how you structure and access data. For example, if a cache entry is n bytes long, you want to ensure your struct is smaller than n bytes. Having very consistent access patterns can also help (e.g. arrays-of-structs vs structs-of-arrays).

This field is super deep, it's very fun to learn about!

Same, from Chrome on Android

That got a chuckle out of me too. When state secrecy and systematic incompetence get in bed together, a lot of things end up being 'decided' lol.

Edit: as a massive fan of industrial and engineering history, the spotty nature of good Soviet documentation does my head in!

Thanks for mentioning the A380 CAD snafu, it made me rewatch one of my favourite clips [0].

The guy taking his hat off makes me burst into nervous giggles every time lol.

[0] https://www.reddit.com/r/interestingasfuck/comments/1k80h9j/...

Damn, that's an interesting point. Having a model transform your work into 'independent' IP and then training on that. I would assume any actual trade secrets (e.g. recipes, financial info) are protected but... tech companies have bent the law before.

I'm assuming they've all at least thought about it quite hard, which is worrying in-and-of itself :).

I think innocent-until-proven-guilty is the correct approach in general but... It's also immature to assume your data will stay private in the long term.

These AI companies are immensely valuable targets. When they get breached, their datasets will inevitably penetrate public datasets. And why would any AI company refuse to train on 'public' data?

Yeah, the thought of having private keys and config slurped up by an LLM provider keeps me away from this kind of direct work.

Perfect job for a local LLM!

It's a tough problem, since I assume any common proof-of-work standard will inevitably generate a big enough dataset to train on.

But nonetheless a problem in desperate need of solving!

These older systems design principles have really scratched a part of my brain and I'm keen to keep pulling that thread. Do you have any recommended readings on the topic?

The thing that always baffles me with vertical farming is sunlight. Assuming most crops are pretty good at turning full spectrum sunlight into useful stuff, why shrink your solar energy per crop?

And assuming you get around this via grow lights, surely the energy and material cost goes up too much for high-volume crops to make economical sense.

This feels about right to me. Living with kids and commuting via public transport (in a country where face masks are not common) might break 5%.

My understanding was that thinking still gets encrypted, shared with clients, and reingested by Anthropic with each new prompt [1]. Which means it would cost more than normal tokens, since it has to be decrypted/encrypted with every transaction.

[1] https://blog.cryptographyengineering.com/2026/05/29/fooling-...

Edit: other comments under this post seem to indicate that thinking tokens are cached on the server side as well? I'm a bit confused.

For more mainstream SBCs, totally agree. However a lot of the niche industry boards (especially new ones) only support Yocto out of the gate. Debian might be "supported" by the SoM/CPU vendor but manually reapplying kernel patches and praying it boots is a PITA.

Cool project. Would be nice to have some unfiltered example photos to gauge the sensor quality w.o. more googling.