There is a huge business opportunity to become the world leader in sustainable data center design. Whoever wins could one day have immense influence over every single data center in the world (or even out of this world, where efficient resource utilization would likely be essential). It’s disappointing to see the United States miss out: US companies and research institutions would actually be extremely competitive in this field with the right government incentives.
Sustainable design for datacenters is orthogonal to success in data centers. That's like saying a local bakery should become a leader in pet adoption - it'll score some feel good points, but is almost counterproductive to the actual mission.
A data center (right now) needs to be built fast, with a large amount of always-on power. Sticking a gas turbine behind the meter in the parking lot is one of the most practical ways to get there.
It’s not just a feel good thing. There are real resource limits, even if someone chooses to foolishly not believe that. If you can get more done with a given level of power and other resources, you get more out of your data centers.
"Given level of power" is meaningless for data centers. If they can get 50MW of power, they'll obviously try to use it as efficiently as possible to generate as much compute/value.
But that's not what anyone means when they're talking about sustainability for data centers in terms of power usage: they're either talking less power (ie less value) or power from renewable sources, which can't compete with sources like gas turbines across a mix of setup time, consistency, and power/$.
If solar and batteries were available sooner than gas turbines, GE wouldn't have a multiyear waiting list.
If solar/batteries were cheaper than gas turbines after accounting for land/permitting/maintenance, every DC would be buying the empty land next to them for an eventual switch to solar.
Notably these many-million dollar investments aren't doing those things, and I guarantee you they have tried every angle imaginable to get power to the DC as quickly and cheaply as possible.
I don’t mean for one data center, I mean globally. There are resource limits to this planet, and we’ve been stretching past them in the short term, and are starting to see the long term consequences. The problem is people think “oh, it’s ok if I pollute a little more because the world is large, and the god we’ll create in our image with these machines will figure it all out for us.” That’s an unsustainable business model based on impressive technology managed by people stuck in childlike fantasies.
I think most people who are working towards greater AI would consider that not an opportunity, but planning to lose in a game that is winner(s) take all.
Winning scenarios in AI don't include being accountable for sustainability and they don't include expertise being valuable. Winners will decide what sustainable means for them.
The situation only has to be sustainable for those entities that get a vote that matters. I don't think they've given us any reason to believe they intend on that being a whole lot of people. In fact they've been quite up-front about saying they are concerned about the possibility of that number being zero
Winner take all is a farce. It didn't work for Uber when they tried to out-spend the competition, Lyft is still very much a thing, and so are taxis (the non-robo kind). There will always be many, many AI companies and trying to execute "winner take all" in this space is only going to make more problems and waste a shit-ton of money.
If you subscribe to an AGI/ASI view, the stakes are different than the network effects that Uber thinks they could get. Success here would mean you could theoretically pull away from the competition so far and so fast, whether that be at a company or nation-state level.
Isn't the goal to power most non-earthbound data centers with solar? I am no scientist, but going out to space feels like it gets rid of most of the downsides to solar.
There once was a time when tech companies mostly stayed out of SF. People who worked in the valley but hated it lived in SF amongst mostly people who had nothing to do with the tech industry. Then someone had the bright idea to put branch offices or even whole companies in SF. Convenient for the commute (and you could move to the East or North Bay), but then SF got overrun by the tech monoculture + NIMBYs trying to “keep it the same” by preventing construction, effectively kicking out so many of the people who were the reason why valleyphobic tech workers sought out the City of Saffron in the first place.
But a language being “typed” doesn’t tell us anything useful. Untyped languages are typed too: they’re uni-typed (every expression is an expression).
I think you do your analysis a disservice by focusing on “is assembly language typed?” as the top line question. The more interesting question you examine is what do the type constraints in inline asm offer, and how do they interact with the host language’s type system?
I know that "untyped" means a single-type, but assembly operands have multiple different kinds of types (as I state in the article). What makes it really interesting is what you can know about each instruction and what it does (what operands it excepts, what it clobbers, what side-effects its has, etc).
And from that huge table of type information, this can be used to give good error messages and suggestions to the user because the compiler actually knows all of this. The type constraints here allow for a lot more than information that normal assemblers just don't give.
In other words, the type of an assembly instruction specifies its effects and coeffects. This is an active research area—describing effects and coeffects in the type system, and discharging handler/provider obligations at the compiler level.
I suspect the main reason someone might quibble over the “assembly is typed” assertion is that many programmers have a rather narrow view of type systems, heavily skewed by OOP patterns.
That's pretty much the quibble. Most people's narrow view of type system.
And we already track all of the basic side-effects and clobbering that each form of each mnemonic does. That's kind of the entire point of this being possible: it's all "typed".
The only shaming I’ve seen are for people using AI badly or not at all.
I also think we should differentiate between programming to get a program you can do something else with, programming to understand the machine, and programming to give the programmer a logical challenge to keep their skills sharp.
The easiest would probably be the smallest plausible set of actual server components in a 5u/tower case, but a custom eATX could go a little further with the cost cutting.
Tried FSD once. It was very confused about a traffic circle, cruised up to it & slammed on the brakes. Got rear ended (at low speed).
The incomplete, miscategorizing and glitching display of other vehicles and obstacles does not give me confidence that the car is able to correctly identify what’s around it, let alone drive accordingly.
Even the lane departure warnings are 90% false positives (but I’d rather be warned than not).
Just driving normally is much less stressful and more fun than supervising the car’s attempt at driving like a teenager.
How long ago was it that you tried FSD? My car (HW4) had a lot difficulty like this in the early days of FSD 11 but over the past thirty months it's gone from "good at highway driving, not very good at navigating city streets and it might try to kill me" to "nearly perfect at everything". Over the past six months, most of my interventions have been to avoid it automatically parking in handicap spaces and minor navigation issues when I know there's a quicker/better route. Since FSD 14.2, I've done 23k miles with 94% of that on Full Self Driving. Longest streak was 930 miles where it took me on a road trip out of state and then back home and I never touched the wheel; it handled parking at chargers and diners, navigating strange destinations (small towns, back roads), avoiding two deer that I almost certainly would have hit, and managed to keep 85mph while navigating traffic during rush-hour in Chicago. I've never owned a product, much less a vehicle, that got so much better over time.
I’m glad to hear it’s gotten better. This was a couple of years ago, when they were pushing the free trials. It’s an earlier Model 3, and I barely drive it.
I’m clearly not a typical driver. I love keeping the camera views all on screen for situational awareness. I don’t use cruise control and rarely use GPS navigation, even back when I used to drive cross country in previous cars. I guess I just got used to memorizing entire routes in my long distance biking days. Google Maps Street View really helps me understand exactly where I’ll be going.
I’m also wondering when this was. I had similar experiences with my first Tesla (2022) and stopped paying for FSD except when I had road trips (then cancel again after). I’ve only felt comfortable using it by default in the past year.
Were Roland MIDI keyboards used by the developers to capture game audio tracks as played by musicians? Maybe they already had the driver code from that.
That's a cogent theory, also, in that I am sure many game composers used Roland hardware. So you were basically, "playing it back non-lossy" as "intended by the composer", in many ways.-
reply