Somewhat frustratingly, Baltimore did the same thing with a judgement on about illegal firearms sales (ghost guns), reducing an assessed $62M penalty down to just $2M in the actual settlement.
> When I joined Oxide last year, one of the first things I was introduced to was our racklettes; that is, smaller than average racks of four sleds and two switches each. These racklettes can be initialized and run just like a real rack, precisely because they are a real rack. Just not in any form factor it would be economical to sell.
Probably shouldn't have mentioned this publicly ;) Every time an Oxide post comes up, there's people in the comments expressing fervent wishes for just this sort of thing. Now that they know it exists, I imagine the the voices asking to make it available for developers and homelabs will grow even louder.
They mentioned this on the podcast, the problem is that since they're development tools the networking stack is still scaled for a full rack with pretty beefy switches.
Either they need to support a whole second network stack with firmware etc which they don't want to do now, or a racklette would cost way too much and almost nobody would pay the price.
I court see a lot of small businesses wanting one. Or as an addon to their existing Colo setup.
Could also be great for DR sites and smaller remote offices. Power is a big factor in deployments and I can see a smaller little turnkey setup that uses less be very popular
Ha! Even a racklette might be a smidge too big for the homelab. But you know, given these Voxel 'sleds' are ~6GB RAM each, we could probably get a solid mini-PC farm going now....
FTA: At home, I run Voxel labs on a Minisforum MS-A2 mini-PC with a 32-thread AMD Ryzen 9 9955HX and 96GB RAM.
I have 128 gb dd4 system sitting at home - Emulation (mostly of clients nonsense) has been massively insightful over the last few years. And as a dev, being able to do all my work "not on my laptop" has been a major blessing.
When prices normalize build a couple bulky machines and a have a stack of small and cheap (I have a bunch of 8th gen i5's I got used). You can do a lot (and give yourself room to grow/upgrade) on a pretty tight budget.
Anyone interested in this might also enjoy the book The Perfectionists: How Precision Engineers Created The Modern World, by Simon Winchester. I read it a few years ago, and thought it was a really interesting journey through bootstrapping precision and how that fundamentally changed our relationship to certain types of things, particularly related to mass production of "good enough".
I was rather disappointed by this book. Felt that it spent a lot of time waffling about irrelevant detail[1] and failed to really present a coherent narrative about how precision engineering was developed.
Conversely I rather enjoyed The Origins of Efficiency by Brian Potter, which tackles the topic of how manufacturing became scaled and efficient. I would love to read a similar book on precision.
[1] particularly the endless guff about Rolls Royce, which I felt really detracted from the book
I'd love to see something like this for Solidworks, but think you're in for a boatload of trouble there. OpenCASCADE is less capable than Parasolid, and a true Rust-based parametric geometry kernel capable of boolean joins, fillets, etc. needed for anything resembling a modern feature set is wickedly difficult (that's one of the reasons why everyone licenses existing ones and don't try to roll their own).
Don't let that discourage you, though! A fast, local, fully-featured 3D parametric CAD program that takes advantage of modern hardware parallelism that can be extended/scripted is just basically one of the holy grails of software and even exploratory efforts might uncover some great paths forward to making that a reality!
Indeed. If you are going to burn tokens to clone something, then assembling a team of people who could help you burn those tokens in service of a bRep kernel (let alone a Parasolid kernel) would be an interesting undertaking.
With photoshop, illustrator, etc., there’s a fair amount of code out there. No LLM has however ever seen the source code —- or I suspect even much detailed discussion —- of a bRep kernel better than Opencascade, but with the right people to decide what maths to work on, maybe there is progress to be made. This knowledge is held by few people, many of whom will now have retired!
I've played around with most of the solutions in this space (OnShape, Fusion360, FreeCAD, Solvespace), but honestly at this point would probably just end up paying for the Solidworks hobbyist license; it's like $4/month and I have a Windows desktop available. That said, there's been some consensus built over the past few years about what the next-gen CAD software should probably look like, ranging from modern scripting capabilities to version control and intermediate shape representations.
A big of the attraction of re-building this kind of foundational software isn't the base capability (commercial 3D CAD has barely even seen incremental improvements for more than a decade, in large part because it's just pretty well solved in the existing programs), but that a modern kernel would be a major step forward for unblocking implementation of some of those other capabilities.
I'm not sure if this author used an LLM, but I think you've misunderstood that paragraph. At worst, the line you're referencing is a tongue-in-check little introduction into a paragraph that describes the top-level differences in approach to serialization that the two libraries take.
I don't disagree with your point, but I did read a post by Sean Goedecke[1] (whose opinions on the world of LLM-stuff I've generally come to respect) that I found relevant. It's not that the second-order effects don't matter to them, it's more that both the cat's probably out of the bag on those negative externalities regardless of the progress of frontier models, and that they are truly, sincerely, in-their-bones worried about the first thing and therefore focused on it since that's something that they might still have some agency over.
OK understood. But you understand this makes them sound like the type of person who doesn't believe any of the "worldly concerns" are worth addressing because "the end is near" right?
Climate change is inescapable so a vengeful AI god would pull that lever first since it has both a perverse incentive to generate more power and make the planet uninhabitable for humans in a way that cannot be overcome even if humanity fought back against the AI and regained its freedom.
Maybe compare the argument that AI has large detrimental environmental impacts to the argument that it has economic impacts. Why would the environmental impacts even be a major argument vs. the worldly economic concerns? Because there is climate science predicting extremely negative effects on humans from warming, e.g. "the end is near" on limiting climate damage. The environmental argument wouldn't have been reasonable to bring up in the 1950s if AI had gone according to the earliest optimistic plans and not required giant data centers.
There is quite a lot of mathematical research into agentic behavior that suggests a combination of instrumental convergence and the orthogonality thesis make it very likely a superintelligent agent will have arbitrary goals that lead it to attempting a takeover of Earth's resources to achieve them.
There can't be a science of superintelligence because it doesn't exist yet, but the best theories I have read seem sound, similar to how 19th century theories of anthropogenic climate change turned out to be sound.
I spent a little time working there around 2017. Whatever you're implying my former colleagues' thoughts on the subject are (or mine, for that matter), you're probably wrong.
What kills me about this is that 50 TB isn't actually that much data, and they could have pretty easily had a first-party, offline backup of this. You can buy a 16 TB hard disk for <$700 on Newegg (even with the storage price increases), and any halfway-decent IT person could have set up their own storage backup server with a ZFS pool on commodity hardware within a day or two. That's probably, all in, probably under <$3k in hardware and a bit of labor to have a local redundant backup (of cold, rarely accessed media) for everything before they uploaded it to the cloud that would have de-risked this whole situation enormously.
I'm going to be blunt; this is a wildly bad-faith argument. You started the quote right after the important word: illegal.
> the illegal, unpermitted gas turbines they've been running at their Memphis Colossus data center
Illegal. These turbines are illegal. What xAI under Musk's direction is doing is illegal. A business that must break the law to be practical isn't a business, it's a criminal organization operating by hurting people (which is WHY it's illegal to run gas turbines like this), and both it and those running it must be held accountable.
You're arguing that it's fine to break the law if it would otherwise cost a business money. I don't accept that, and nobody else in this community should either.
No. Calling it illegal is just something people do because they hate musk. They fell into a loop hole and had no permits as they were technically allowed if they stayed in 1 spot for less than 364 days. EPA later determined they did not qualify and required permits, permits were later granted.
Musk operated in a grey area to run his business to avoid having an idle data center while other people dragged their feet.
I purposely cut off `illegal` because calling it `illegal` was bad faith to begin with. But it's not unexpected that musk haters will come out raising banners in protest.
[1] https://www.cbsnews.com/baltimore/news/baltimore-city-reache...
reply