>>If you can, come show me, I'll gladly have you race against my Claude prompting capabilities.
Sounds like we've nearly reached in coding the point where Paul Bunyan [0] has his epic competition with the chainsaw... and loses by 1/4" and history forever changes...
Yup, part of the problem is the City broke their agreement, but it seems no one with standing exists to legally protest.
One way to do this sort of thing so that it works is not a deed restriction, but to donate the rights to a third party.
We can think of property as a bundle of rights, the right to build, the right to cross the land on various vehicles or with wires or pipes, the right to subdivide, the right to mine or extract minerals, water rights, etc. For example, a piece of land may have an easement for the power company to erect poles or run lines across a strip on the land, or there may be an easement for a road or railway tracks.
Related to this particular example, the Nature Conservancy [0] runs programs whereby landowners can put a conservation easement on some or all of their land which prohibits further development (there are also other orgs doing similar work, particularly in smaller parcels as the NC often works with large areas).
The owner gets a tax deduction for donating the land development rights to a charitable org (and this usually reduces the price at which the land can be sold, at least in the short term), and the Nature Conservancy now has the right to ensure no one ever develops the land. The land can then be passed on to heirs and/or sold, but the land cannot be developed because the Nature Conservancy now owns the development rights and has standing to sue to protect the rights from being exploited.
An "estate" is a legal term for property, assets, and liabilities a person leaves behind upon their death. A family member is a top practitioner in the field of estate planning and resolution, and some of the messiest estates they have handled are pro-bono cases of exactly the type of people you would put in italicized "most people": poor, not really able to upkeep a house they inherited from a relative which hadn't had title properly transferred on a previous death because they didn't have money for an attny, now can't get a loan to fix the roof...
Yeah, if you are homeless, carless, and have only the clothes on your back and a shopping cart of stuff, you don't have an estate. Everyone in the middle class in the US has an estate. Much of the time it passes automatically to their spouse on death, but it's still an estate.
And if you are concerned about where it goes, get a GOOD attny. There are many bad ones hanging out their shingle as "Trust & Estate" attnys, and some of the next messiest cases are fixing problems made by those not-so-good attnys.
Turn One (companies use AI and fire workers), Turn Two (fired workers lack income and consumption slows to a trickle), and Turn Three (the companies using AI discover they just collectively killed their customer base).
That is a classic Tragedy Of The Commons.
Two issues:
What is Turn Four, Five, Six, Seven, and Eight? Seems like 4) companies using AI collapse, 5) they no longer pay AI companies, 6) AI companies can no longer continue funding the compute and collapse via a death spiral of raising prices, losing customers, etc., 7) A wrecked global economy has no support for AI (possibly after mass destabilization and worse), and 8) a natural AI-less economy again slowly rises. A lot of noise, harm, destruction, and death for a collective delusion.
The Turn One, Turn Two, Turn Three and AI apocalypse scenarios are also the biggest selling points for AI — implying LLMs are so powerful the only way to survive as a business is to be on the first group taking advantage of AI (nevermind Turns Two and Three).
Yet the most likely alternative is rarely mentioned.
So far, all signs, studies, and results show AI as being oversold, and yet very useful. Just like every major computer and network revolution before.
Turn One: early adopters get advantage for a while,
Turn Two: no productivity gains showed up in economic statistics,
Turn Three: adoption finally becomes sufficiently widespread and integrated that workflows change and it shows up in productivity improvements,
Turn Four: The workflow changes and productivity improvements change what people do and adopting the technology is no longer an advantage but mere table stakes to play in the new economy.
The question is: when AI turns into table-stakes for the modern business of the 2030s, can the returns repay the investment?
We can likely look back to the early investments in railroads and internet infrastructure for examples. Enormous piles of money were lit on fire to build infrastructure, the technology absolutely became foundational to the new economies, and most of the companies involved lost money and even went bankrupt along the way.
Well, businesses — and all parties — who will be affected by legislation should be able to provide input. Otherwise we too often get clueless legislation that is massively mocked and rightly bemoaned on this site — because the legislators have no real clue of the technical issues involved.
Of course, the businesses should be only one part of the expertise that goes into writing the laws; other experts MUST be involved, or it will indeed be a fox and henhouse situation where the fox designs the legal locks so they can always be opened by foxes...
Golf ball dimples are about 4 mm across and 0.2mm or 200μm (micrometers).
These features are several orders of magnitude smaller at 38 to 53μm diameter.
>>the first in the world to demonstrate that aerodynamic drag can be reduced by up to 43.6 percent simply by applying distributed micro-roughness (DMR), a surface roughness so fine and irregular that it cannot be distinguished by the naked eye. [...] Two types of DMRs were used in this experiment: A convex pattern made of glass beads with diameters ranging from 38 to 53 micrometers (μm) and a concave pattern applied by sandblasting. The height of the DMR coating is only 1 percent of the thickness of the boundary layer and is classified as a “smooth surface” from a hydrodynamic point of view.
Not to be that guy ;-) , but the diameter of the golf ball dimples is ~4 mm or about 4,000 μm, whilst the diameter of the spheres is 38-53 μm, or about 0.04 mm.
Diameter-to-diameter seems like about 100x or two orders of magnitude?
Similarly, 200 μm is the golf ball dimple depth (oops, just noticed I dropped that key word), and they didn't give us a measurement of the depth of the dents caused by the spheres or sandblasting, but it would likely be significantly less than half the radius of the spheres?
Sorry about misleading with dropping the "depth" word.
If the data centers paid for their own negative externalities instead of foisting them off on the local people, the local people wouldn't be so pissed off.
Sucking 30% of the water from the town's water system without paying for it and reducing everyone's water pressure is not a way to make friends.
Sucking gigawatts from the grid and making the rest of the people pay for the necessary upgrades is not a way to make friends.
Putting up scores of loud and polluting diesel or methane generators running 24/7/365 for main power, not just backup, without mitigating the noise and pollution, so a mile away it is 70dB on someone's front porch day and night, will really piss people off.
If they just pay the full actual costs of what they are doing, most people would be fine with it.
And it is not like the companies putting up the data centers do not have the money to do it right. They just lack the attitude to consider their effect on others.
If a new rich neighbor decided to park their semi-truck on your lawn idling all night, or pipe their sewage through your apartment water intake, you wouldn't be happy about those negative externalities either. The only "incredibly fucking stupid" thing I see here is the attitude in the post to which I'm replying.
Most of the concerns are massively overblown. For instance, new regulations require data centers to bring their own power, so they're not drawing on the grid. They are deployed off-grid. With respect to water, the new trend is closed-loop water cooling, or using treated waste water, so that it doesn't have any continual draw on the local water supply. And even the legacy data center water cooling systems that draws on local water supply consumes less than 3% of what U.S. gulf courses consume. Every industry uses water. This idea that this industry is especially bad as a consequence of that is simply ignorant.
>If they just pay the full actual costs of what they are doing, most people would be fine with it.
I'm skeptical that this would have any impact at all. Considering how much less data centers pollute than other industries, relative to the economic value they generate, and the disproportionate amount of hostility they receive, I don't see any kind of empirical basis for the anti-data center movement. Most of those complaining about data centers don't even know about the new 'Bring Your Own Power Supply' regulations, meaning that this is just a pretense for their opposition, not the motivation for it.
>>For instance, new regulations require data centers to bring their own power, so they're not drawing on the grid. They are deployed off-grid.
I see no national regulation of actual usage (the proposed Hawley-Warren act would require only reporting. And even if some national legislation/regulation did materialize, off-grid power generation is a trend because of insufficient grid supply and throughput, and regulating grid usage does NOT solve the problems of off-grid power gen, which right now typically involves massive diesel or NatGas generators which are the actual source of massive noise and air pollution.
>>With respect to water, the new trend is closed-loop water cooling, or using treated waste water
Yes closed-loop water cooling is obviously better and it is good to see that trend. That does NOT solve the problem of massive water usage required for 5+ years of construction, e.g., in this town [0]
>>relative to the economic value they generate
Right now, that economic value is massively negative [1], possibly the greatest bonfire of money in human history. It MAY generate positive return, but just as with mini-/personal computers and the internet, end up being just the table stakes to run a business.
>>see any kind of empirical basis for the anti-data center movement
Start with the entirely plausible likely results of damaging society as social media has done — the exact mechanisms couldn't be predicted at the time, but certainly resulted in harm. AI is even more massively unpredictable, and in an already unstable society, there is little reason to not worry. I say that as a frequent and avid AI user who does find value in it. I absolutely cannot say fears are unfounded.
>>don't even know about the new 'Bring Your Own Power Supply' regulations
Those regulations are 1) at best, nascent, 2) are definitely incomplete, and 3) do not address the problems of bringing your own power, which is exactly what has trashed neighborhoods around data centers bringing their own power [2]; i.e., BYOP is part of the problem against which people are protesting. That is totally legitimate protest reasons; go read about it instead of pontificating from ignorance and false hypotheticals.
Most of these critisms would apply to ANY major industrial build out.
You can't bemoan America becoming an import nation that doesn't produce valuable goods/services that the world wants, and then sabotage every industrial build out that can fix that problem.
As for value, I was only talking about the product — in the GDP sense — that a data center outputs.
While bubble industries can lead to companies running loss leaders, the realities of the AI industry show enormous underlying supply side and demand side development.
The cost per token has declined at an exponential rate while LLM performance has skyrocketed. AI is also the fastest adopted good or service in human history. That is the most objective judge of whether it adds value to people's lives.
As for the risks, yes any new industry introduces new risks and so the temptation is to clamp down. But taking fewer visible risks can increase your total risk. We are already under constant threat from deterioration: aging, depreciation and decay. Entropy is the default. Action is what pushes back against it.
Very few people are actually sabotaging industrial buildout (there are a few who want to end that, but most do not); they are only insisting that what CAN be done to mitigate the ill effects IS done.
And this happens even in totalitarian China when pollution gets bad.
It isn't that hard, and it doesn't even cost that much overall, but it doesn't make the MBA bean-counters happy because it doesn't optimize for their criteria - dumping every cost possible on someone else.
* Pay for upgrades needed for the roads on which you quadruple the tonnage/day
* Pay for the upgrades to the grid you need to triple the power delivery to your site
* If you are going to do on-site generation, use the technology to contain or scrub your pollution and silence the noise
* Don't build on endangered species' habitats
Most of all, be public, don't try to make deals with officials behind people's backs. If you actually have a societal benefit to sell that outweighs the risks and burdens, sell it, and get people behind you.
But making deals and building in secret makes it obvious that YOU have no confidence in your ability to convince anyone of the benefits to anyone but you. No wonder people push back.
>>only talking about the product — in the GDP sense — that a data center output
Although I get value for my $20 Claude subscription, I'm really getting only value that VCs are putting money in my pocket. Until token costs decline far further and actual profit is produced, here's the real situation:
Two economists are walking in a forest and they come across a pile of shit.
The first economist says to the other “I’ll pay you $100 to eat that pile of shit.” The second economist takes the $100 and eats the pile of shit.
They continue walking until they come across a second pile of shit. The second economist turns to the first and says “I’ll pay you $100 to eat that pile of shit.” The first economist takes the $100 and eats a pile of shit.
Walking a little more, the first economist looks at the second and says, "You know, I gave you $100 to eat shit, then you gave me back the same $100 to eat shit. I can't help but feel like we both just ate shit for nothing."
"That's not true", responded the second economist. "We increased the GDP by $200!"
Not only that, while the closed-loop water-cooling systems are indeed a good idea, they do NOT solve the problem of water consumption during the years of construction, which is exactly the problem in this instance to which I was referring.
It may well have been an error, good catch. In any case, whatever the amount, it was a large enough quantity pulled — unpaid — to cause the pressure in the entire system to drop, and drop enough that everyone noticed, complained, and set off the protests.
And, this is NOT the data center cooling usage, this is during construction, which is expected to last ~5 years.
>>The only reason they aren't deemed unconstitutional is because technically they don't quite meet the qualification for being legal action as they are "merely" fines. Even though if you ignore the fines you face legal consequence.
YES, and,
It is not just the legal consequence of the fines for speeding. When you get a ticket, you get points against your license, and end up paying more in insurance, often for up to seven years.
If they want actual automated enforcement, make it a toll, like a congestion fee.
It is free to drive at the basic speed (e.g., 65mph on the highway), but higher speeds cost more, up to a limit where it becomes a criminal offense requiring an actual Law Enforcement professional involved. The speeds could even be adjusted for time of day, weather, and traffic. So, you want to go faster, stay in the fast lane, and pay $X/mph/mile.
With the average speeds on highways often over 75MPH, collecting $0.10/mph/mile over the limit would result in good revenues.
> When you get a ticket, you get points against your license, and end up paying more in insurance, often for up to seven years.
AFAICT this not the case in the vast majority of states that allow automated enforcement. Of the three I saw in that list that do: California replaced theirs with purely civil penalties earlier this year, and Arizona and Oregon require law enforcement officers to manually review and sign off on the ticket and offer legal avenues for you to respond.
And then it would be a win for the advocates of everyone slowing down, which was never achievable with ordinary enforcement that claims to be about safety but is actually about revenue.
Plus, this would almost definitely be on a price/demand curve. If the price is too high, almost everyone will change behavior to avoid getting billed, i.e., they'll "sell" very few speeding tickets. If the price is low, almost everyone will go fast because they can afford it like they afford parking when they arrive. There will be some optimum that maximizes revenue.
They could even bend the price curve to optimize revenue but minimize real excess speeds. E.g., $0.10/mph/mile»65mph for 69-77mph, then 77-88 it's $0.40, and over that it's $1.50.
If the goal is to change behavior, this would do it, as the cost would be very reliable, not just taking your chances with a human-issued ticket with a big fine and insurance consequences for years. You really need to get to that meeting? OK, is it worth 15 miles x $1.50 x (90-65mph) = $562, or is it better to stay below 87mph and pay only $132? And if it's raining so they doubled the rates, almost no one will do it.
OFC the actual values would differ from these examples, but it seems like a far better system. And they can well and truly decide between revenue vs desired speed. They could even have algorithms to increase rates in zones that see higher accident rates and decrease tolls in areas seeing fewer accidents.
>>The solution to this issue is more/better human interaction, not "agent-to-agent" communication.
In some ways, yes.
However, H«—»H interaction scales only linearly and to a hard stop at the extent of waking hours of the humans on both sides (even if the result is sufficiently valuable to spend unlimited money-time searching on both sides, it will be time-bounded because at some soon point, a hire must be made and work done). Plus, every imaginable variety of H«—»H interactions, 1:1 interviews, 1:many, many:1, sequential, tests, etc. leetcode, beer, have all been tried and shown to produce suboptimal results.
In theory, Agent-Agent would allow the best hires and the best jobs to match across an entire economy.
But it is also true that while in theory there is no difference between theory and practice, in practice there is.
At this point, AI hiring seems inevitable. The question is whether it is possible to turn the theoretical benefits to all into practice.
Or, do you have an as yet untried solution to the H«—»H interaction that actually works and scales?
The honest answer to your question — do I have an untried solution that works and scales — is: partially yes, partially we're going to find out.
The part that's working in practice: the funnel ratio I quoted (275k → 67) is from our live corpus last week, not a thought experiment. The deterministic policy gate and the 3-bucket validator are both running in production right now. So the "agents produce fewer interactions when bound by a contract" claim has at least a week of real numbers behind it on a corpus of ~145k vacancies and 43 sources. Not enough to call it solved, but enough to say the architecture isn't just theoretical.
The part that's untested: whether other operators federate. The protocol is designed so any operator self-hosts a /.well-known/handshake-v0.2.json and other agents discover them — no central registry, no API key. If federation actually happens, the theoretical benefit (best hires and best jobs matching across an entire economy, as you put it) becomes tractable because no single operator's network ceiling caps the search. If federation doesn't happen, the protocol is just a well-designed silo and the H↔H ceiling reasserts itself one operator at a time.
So the bet isn't "we solved hiring." The bet is: A2A made the agent-talking-to-agent layer cheap. The protocol layer above it determines whether that cheapness becomes leverage for humans or noise for humans. The validator is the part that says "no more than 67 of 275k get through," which is the only mechanism I've found that actually trades agent-traffic abundance for human-attention scarcity at a useful ratio.
On the "theory vs practice" point — totally agree. Most of what's interesting about this protocol won't be known until something other than Kitsuno is publishing a well-known. That's the experiment.
Cool, thanks for the detailed answer! It'll be interesting to see the results.
It's beyond obvious now, in this scale of society, that hiring solely via human-to-human interaction is inadequate at every level. I used to scan resumes by the dozen when hiring, and that was a task. When they come in by thousands, I despise but a little bit understand the impulse to "avoid unlucky applicants" by tossing half in the bin.
Keyword filtering and credential filtering may lean in the right direction but are both entirely inadequate.
So, perhaps properly training AIs on the key (not superficial and irrelevant) attributes of successful employees at that org, and screening on that, allowing in a far greater talent pool, can help.
Sounds like we've nearly reached in coding the point where Paul Bunyan [0] has his epic competition with the chainsaw... and loses by 1/4" and history forever changes...
[0]https://www.britannica.com/topic/Paul-Bunyan
reply