I'm in the same boat and there's an ever increasing number of things that I can't do now thanks to photo ID verification requirements. Want to apply for a bank account online? Point your smartphone at your passport or driving license. Digital signature for a bank transfer? We only support this one platform and guess what they require as verification. Want to gamble? Sorry, photo ID holders only. Hell, you want to apply for a provisional driving license they want your passport, want to apply for a passport they want your driving license.

Yes, there's workarounds but god damn do they make it awkward. I'm just going to have to play their game and go for the in-person checks they do for passports. I know someone who did it recently (as opposed to all those unhelpful people who did it when it was a form at the post office and can't understand that things do in fact get worse over time) and he said it was easy enough, but it did require going outside and speaking with bureaucrats and I'm less than enthused about that.

Which would explain why the AI would generate such code. It learned from the best.

There was more wrong with Eternal than just the difficulty. A *lot* more. Adding in optional *additional* difficulty is fine - pleases the small number of people who want Ultra Nightmare to be harder without taking too much effort. Dark Ages walks back some of the mistakes of Eternal which likely helped it sell, but not nearly enough for my liking.

Depends what you're doing. Rushing to the assembly was always stupid. But it was never hard to outdo a compiler, just usually not worth the time penalty. When working in non-x86 over the years the penalty from the C version of popular tools not containing all the x86 specific optimisations was always massive. Thanks to the popularity of the M series CPUs several big open source programs now have arm native simd assembly backing up their formerly C only fallbacks and it makes things usuable where before they were unusable. I'm eternally grateful that there are still people who understand that C/C++/Rust/etc. just don't have the syntax to describe code in ways that can be optimised to GOOD vector code and are willing to break out the machine specific intrinsics so we can encode video in reasonable timeframes.

As a filthy mac user I tend to find from experience the existence of a java application means it is 100% completely and solidly locked to windows. Not being a Java programmer I don't know why this happens. I suspect it's because the cross-platform nature of the beast runs out of steam as soon as anything non-trivially gui related needs doing and then it's straight to the platform native windows api wrapper to achieve something, and once you've popped that cork, time to let the UNC paths and powershell widget wine flow.

When AI is doing stuff we don't have traditional algorithms to achieve, you can argue the whole-output-of-a-mini-nuke-plant requirements are acceptable. But then people start using it to solve problems we do have efficient solutions for because a programmer capable of understanding the problem and coding efficiently for low spec low power hardware is expensive, but a few hundred hours on a cloud service is "cheap" and fast provided someone else foots the bill in the hope of a "your problem now" payday later. Just because the weights are being processed on dedicated silicon engineered to efficiently do this doesn't absolve them of the great wastage happening as people try to find a practical application (and by practical I really mean profitable) of it before the money runs out.

Market size matters too. Cost of manufacturing and distribution isn't as much as you'd think. Most of the cost is in the development, admin, marketing, exec bonuses, etc. and those costs have gone WAY up so you'd expect games to have shot up in price long before now. But for the most part the larger install base of consoles+PC compared to the old days means you can afford to make less per unit if you're selling a LOT more than your average SNES cart.

Lectures from the big universities turn up on youtube from time to time. Paying your way into the "good" schools clearly doesn't help you dodge time wasting filler. A significant portion of the "lesson" involved reading the man pages from overhead projector slides. Even in my day the understanding was you weren't going to learn a damn thing at university. The goal was to "network" and from speaking to employers at the time, if you hung around for the full duration well that was a red flag. But attitudes to institutions change with the generations. I just happened to be immediately following the big post-dotcom Batchelors in this that and the other cook-off that made companies damn wary about hiring anyone who thinks their expensive education was worth anything.

I do pay for crossover but I frequently wonder why. Whenever I have a windows program I need to run, crossover has only very rarely been the tool that got the job done. I always wind up having to fire up Parallels. And the thing is, while obviously WINE is a hell of an undertaking to get 100% compatibility and I accept that, it's even the native bits that Crossover provide as their value add that just plain suck rocks. The GUI is basic but barely functional. Icons persist for programs long uninstalled, new things installed through their wizard don't appear, and solving issues is a very manual affair. Crashes are extremely common and Steam seems to require a reinstall every few months. Compared to proton which seems to work more than it doesn't, I find Crossover quite poor in practice.

In the pre-Musk days I'd read more than a few accounts of people without checkmarks who didn't really desire one being solicited by twitter employees who could get them one for just a measly wad of cash. When Musk announced that it was just going to be a fee and everyone was angry about that, I remembered the articles about how unless you were a celebrity it was basically already a fee, just a bribe more than a process. Of course, this is me reading journalist's articles during the time when "journalist" was a person who copy/pastes tweets so I suppose it's foolish to assume they did their due diligence, but I can't help but think "no smoke without fire" and all systems are corrupt whenever there's people involved.

I don't remember which version it was, 3.5? It was a long time ago anyway. I had an extremely basic setup on a nice chunky Dell server and I found that while everything worked, the management interface was REALLY clunky and bothersome. It all behaved like an early web app framework where seemed like it was trying to make AJAX requests and pretend it was a real GUI where panels would just get stale data, fail to update automatically and a refresh would cure most ills. Nothing really went wrong as such with the business end of the VMs, but still I had no confidence in the damn thing. Fortunately as a person who just needed about 10 VMs and no SANs or redundant anything in the budget, I just handrolled some kvm based replacements and I've been happy with that ever since. And from what I've seen from KVM, I'm confident that if I ever needed a big boy setup it would be possible, even if I personally might run out of talent. And so I just can't see myself climbing on the ESXi horse again.

VMware Fusion went free some time ago and they don't make it particularly easy. You get stuck in a loop trying to navigate to a page that will let you either download or register to download. Last time I had better luck googling the thing to get me the actual product page. And then you have to have it available on your vmware account which sometimes gets stuck in a "verification" step that never passes or fails. And what's especially weird is that it's not as if they are even trying to make it so difficult you just give up and pay for pro because as a fee paying pro user (on the work account) they broke that too. It also kept trying to get me to log in to my broadcom enterprise account which I don't have and fusion isn't even in that if I did (as I understand it.) I even tried installing an older version hoping the auto-upgrade would sort it out but it just opened safari and took me to a dead landing page. In the end I found a download on archive.org that passed the hash check so I took the risk that it was safe and not some freaky hash collision thing. And after all that the package I was trying to build wouldn't build on debian arm. An evening of failure indeed.

I don't know how real enterprises manage, because every time I need to deal with enterprise portals it's just a complete clusterfuck. One time when I was blessed enough to have a Dell account manager I asked why their website was so awful and he said "to make you call us so we can upsell everything."

I hate gas. I hate the idea of gas, I would like to be rid of it. But I don't have the money lying around to squander on heat pumps. Taking on multi-year debt to retrofit my house for a non-existent saving on the bills wouldn't make any sense when not spending that money will result in a much larger saving from not paying debt interest. And I suspect just like solar, heat pumps will be a detriment to selling your house and not an investment. There's a lot of things in life that are good but the general population believes otherwise and heat pumps may have been in that category at one point. But the rising cost of energy has seen fit to close that gap entirely and now you'd be worse off until some of these mythical cheap renewables actually start bringing down the bills. No, I won't risk what little financial wellbeing I have for the "good" feeling of helping meet a carbon target, and I won't gamble on it paying for itself down the line. In my experience these things rarely do. A few years down the line a new thing will come along that will promise all the things the last one would do, except for real this time. And the only winners are the people who chose not to play.

I'm tired of listening to evangelists for green solutions that didn't pay retail, got subsidies that aren't available to me, scadged installation from their work, charged the expenses to their employer or simply have so much bloody money they don't mind the increased expense, it's just buried in the black budget of their expendable cash and disposable lump sums that came from sources I could never dream of. But that's youtube and the greater internet. "This thing I got from a sponsor is amazing! Please pay no attention to the video from 2 years ago from the other company that no longer exists and I can now be honest about all the problems I had and wouldn't admit to at the time. But THIS one, THIS is the boy!" (For now.)

Running the model (inferencing? not sure of the terminology) is not the "we'll need nukes" thing, even if it is technically much more power hungry than a non-LLM solution (if such a thing exists.) Generating the models in the first place, that's the power hog where you task an entire datacentre full of power guzzling GPUs for months to make a few hundred gig or more of weights that then "only" peg your consumer CPU/GPU when quantised enough to fit in your RAM for a few seconds.

People who have to work in environments without heavyweight IDEs and consistent settings will never accept whitespace sensitive languages. For us it's a non-starter because there's literally no way we can have something break because of an invisible character that was added/removed by an errant editor setting someone in the chain didn't know to check for before making a "quick change." And so when the solution to a problem is proposed in python, we will naturally get the yips because of past experience. At this point I'm less suspicious of a .php than a .py. PHP was a useful language with a design that let terrible people do terrible things. Python seems to be a terrible thing that causes terrible things to happen under the guise of preventing sloppy code.