An anonymous reader quotes a report from CleanTechnica: Spanish renewable energy giant and offshore wind energy leader Siemens Gamesa Renewable Energy last week inaugurated operations of its electrothermal energy storage system which can store up to 130 megawatt-hours of electricity for a week in volcanic rock. The newly-opened electric thermal energy storage system is billed by Siemens Gamesa as "The Future Energy Solution" and as costing "significantly" less than classic energy storage solutions. Specifically, according to the company, even at the gigawatt-hour (GWh) pilot scale, ETES "would be highly competitive compared to other available storage technologies."

The heat storage facility consists of around 1,000 tonnes of volcanic rock which is used as the storage medium. The rock is fed with electrical energy which is then converted into hot air by means of a resistance heater and a blower that, in turn, heats the rock to 750C/1382F. When demand requires the stored energy, ETES uses a steam turbine to re-electrify the stored energy and feeds it back into the grid. The new ETES facility in Hamburg-Altenwerder can store up to 130 MWh of thermal energy for a week, and storage capacity remains constant throughout the charging cycles.

An anonymous reader quotes a report from VentureBeat: At the International Supercomputing Conference (ISC) in Frankfurt, Germany this week, Santa Clara-based chipmaker Nvidia announced that it will support processors architected by British semiconductor design company Arm. Nvidia anticipates that the partnership will pave the way for supercomputers capable of "exascale" performance -- in other words, of completing at least a quintillion floating point computations ("flops") per second, where a flop equals two 15-digit numbers multiplied together. Nvidia says that by 2020 it will contribute its full stack of AI and high-performance computing (HPC) software to the Arm ecosystem, which by Nvidia's estimation now accelerates over 600 HPC applications and machine learning frameworks. Among other resources and services, it will make available CUDA-X libraries, graphics-accelerated frameworks, software development kits, PGI compilers with OpenACC support, and profilers. Nvidia founder and CEO Jensen Huang pointed out in a statement that, thanks to this commitment, Nvidia will soon accelerate all major processor architectures: x86, IBM's Power, and Arm. "As traditional compute scaling has ended, the world's supercomputers have become power constrained," said Huang. "Our support for Arm, which designs the world's most energy-efficient CPU architecture, is a giant step forward that builds on initiatives Nvidia is driving to provide the HPC industry a more power-efficient future."

"The U.S. military's Cyber Command has gotten more aggressive than ever against Russia in the past year, placing 'potentially crippling malware' in systems that control the country's electrical grid," according to CNET, citing a report in the New York Times: Made possible by little-noticed legal authority granted last summer by Congress, Cyber Command's strategy shift from a defensive to offensive posture is meant in part as a warning shot, but it's also designed to enable paralysing cyberattacks in the event of a conflict, The New York Times said Saturday, quoting unnamed officials... [T]he recent moves appear to have taken place under a military authorization bill Congress passed in 2018 that gives the go-ahead for "clandestine military activity" in cyberspace to "deter, safeguard or defend against attacks or malicious cyberactivities against the United States...."

The Times said Cyber Command is concerned Russia could trigger selective power outages in key states during the 2020 election and that it needs a way to discourage such attacks. But the agency and the U.S. have to consider their moves carefully in this international game of cyberchess. "The question now is whether placing the equivalent of land mines in a foreign power network is the right way to deter Russia," the Times report says. "While it parallels Cold War nuclear strategy, it also enshrines power grids as a legitimate target...."

In related news, Bloomberg reported Friday that a Russia-linked hacking group that shut down an oil and gas facility in Saudi Arabia in 2017 has been probing utilities in the U.S. since late last year.

An anonymous reader quotes Computerworld columnist Mike Elgan: Ambient computing is real. It's the next megatrend in computing.... To interact in an "ambient computing" context means to not care and not even necessarily know where exactly the devices are that you're interacting with. When IoT devices and sensors are all around us, and artificial intelligence can understand human contexts for what's happening and act accordingly and in our interests, then ambient computing will have arrived...

As with many technology revolutions, including augmented reality and AI, the buzzword ambient will precede the actual technology by many years. In fact, the marketing buzzword is suddenly here in full force. The actual technologies? Not so much. Instead, we're on the brink of a revolution in what you might call "semi-ambient computing...."

Rumors are circulating that Google's next smartphones, the Pixel 4 line, may come with Soli built in. I told you in January about Google's Project Soli, which may be called the "Aware" sensor or feature in the Pixel 4 -- again, according to unconfirmed rumors. Soli or Aware capability means the Pixel 4 may accept in-the-air hand gestures, such as "skip" and "silence" during music playback. The new Google "wave" is a hand gesture. The ability to wave away music with a hand gesture brings the smartphone into the semi-ambient computing era. It basically adds natural hand gestures to natural-language processing.... Google also briefly talked last year about a healthcare assistant called Dr. Liz., which was described by former Google CEO Eric Schmidt as an ambient computing virtual assistant for doctors. We'll see if Google ever ships a Dr. Liz product...

Yes, ambient computing is real, and the Next Big Thing, showing up first in business, enterprises and healthcare. But for now, the term ambient computing will be misapplied. It's a buzzword that will be stapled to every semi-ambient computing product and service that comes out over the next few years.
The article predicts we'll eventually see ambient computing arriving in cars, grocery stores, smart glasses -- and notes a Microsoft job listing for its "Ambient Computing & Robotics team" describing "the era where computer vision, AI-based cognition, and autonomous electro-mechanicals pervade the workplace."

Computerworldd adds that Microsoft "was mocked for its 'Clippy' assistant, which the company released in 1996 as a way to provide friendly help for people using Microsoft Office. In the future, Microsoft may release what will essentially be a Clippy that works, because it will understand human context through AI."

A reader quotes the BBC: A massive electrical failure has left almost all of Argentina and Uruguay without power, according to a major Argentine electricity provider. Authorities say the cause of the blackout is still unclear. Argentine media said the power cut occurred shortly after 07:00 [03:00 PST, 11:00 BST], causing trains to be halted and failures with traffic signalling.

It came as people in parts of Argentina were preparing to go to the polls for local elections.

"A massive failure in the electrical interconnection system left all of Argentina and Uruguay without power," electricity supply company Edesur said in a tweet. Alejandra Martinez, a spokeswoman for the company, described the power cut as unprecedented. "This is the first time something like this has happened across the entire country." Argentina's energy secretary, Gustavo Lopetegui, said the cause of the power failure had not yet been determined. The Ministry of Civil Protection estimated that parts of the service could be restored in about seven or eight hours.

Edesur said that power had been restored over 75,00 clients in parts of Buenos Aires and local media reported that two airports were operating on generators in the capital. Uruguay's energy company, UTE, said in a series of tweets that power had been restored to coastal areas and to areas north of Rio Negro.

The combined population of Argentina and Uruguay is about 48 million people.... Tierra del Fuego in the far south is the only area that remains unaffected because it is not connected to the power grid.
"Local media have been showing voters casting their ballots in the dark, with mobile phones being used as lanterns."

AMD is preparing a monstrous 64-core/128-thread Threadripper CPU for launch in Q4 2019. "AMD's largest HEDT processor right now is the W2990X which tops out at 32-cores," reports Wccftech. "This is nothing to sneeze at and is already the highest core HEDT part around but because the world can't get enough of these yummy cores, AMD is planning to launch a 64-core version in Q4 2019." From the report: The platform is called X599 right now although I am told AMD is considering changing the name to avoid confusion with Intel. This is not really surprising since both Intel and AMD HEDT platforms have the same nomenclature and it can get really confusing. I am also told that they they plan to retain the "99" suffix. AMD is planning to launch the 64-core Threadripper part and the corresponding platform in Q4 2019. In fact, that is when you can expect these motherboards to start popping up from various AIBs.

Now my source did not mention a new socket, so as far as I know, this should be socket compatible with the existing TR4 motherboards and only a bios update should be needed if you already own one. What I don't know right now is whether this is a 14nm part or a 7nm part. Conventional wisdom would dictate that this is a 14nm part trickling down from their server space, but who knows, maybe the company will surprise all of us? This is pretty exciting news, because knowing AMD, the 64-core Threadripper CPU will probably be priced in the $2500 to $3000 range, making it one of the most affordable workstation processors around with this many threads.

Catalyst is Apple's framework that enables developers to easily bring existing iOS apps to the Mac, while SwiftUI is a new, Swift-based technology that makes it easy for developers to create one app that runs on all of Apple's platforms. Jason Snell from Macworld highlights the slow, invisible transition of these new technologies. From the report: Catalyst, which arrives this fall, will allow developers who are well-versed in the vagaries of writing iOS apps to use those skills to write Mac apps. This will most commonly take the form of bringing iPad apps to the Mac, with additions to make them feel more like native Mac apps, but it's more than that -- it provides iOS developers with a familiar set of tools and access to an entirely new platform, and it makes the target for professional apps across Apple's platforms broader by including both the iPad and the Mac. iOS apps are currently built to run on devices running Apple-designed ARM processors, and if the rumors are true, that's another transition waiting to happen. But given that all Mac and iOS developers are already using Apple's Xcode tools to develop their apps, I suspect that the pieces have been put in place for a fairly simple transition to a new processor architecture.

And then there's SwiftUI, which may be a harder concept for regular users to grasp, but it's a huge step on Apple's part. This is Apple's ultimate long game -- an entirely new way to design and build apps across all of Apple's platforms, based on the Swift language (introduced five years ago as yet another part of Apple's long game). In the shorter term, iOS app developers will be able to reach to the Mac via Catalyst. But in the longer term, Apple is creating a new, unified development approach to all of Apple's devices, based in Swift and SwiftUI. Viewed from this perspective, Catalyst feels more like a transitional technology than the future of Apple's platforms. But we're talking about the long game here. Transitional technologies are all a part of the long game. Catalyst will bring those apps to the Mac. iOS and Mac developers will pick up Swift and SwiftUI. Mac apps can integrate iOS stuff via Catalyst. iOS apps can integrate Mac stuff for use on the Mac. And all developers can begin experimenting with SwiftUI, building new interfaces and replacing old ones in a gradual process. "And then we'll turn around sometime in the 2020s and realize that all of this talk of UIKit and AppKit and Catalyst is behind us, and that our apps are written in Swift with interfaces created using SwiftUI," Snell writes in closing. "It will have all changed due to Apple's slow and steady pace of iterative, continuous improvement. The long game never stops, and it can be hard to see that you're even in it."

The System76 Gazelle laptop is finally available. From a report: What makes this laptop so special (besides its pre-installed Linux-based operating System), is its impressive specifications. You see, System76 has supercharged it with a 9th generation Intel Core i7 Processor (9750H) and NVIDIA GTX 16-Series Graphics. It even has something the pricey MacBook Pro doesn't -- the ability to be configure with up to 64GB. Yeah, Apple's laptop can only have a maximum of 32GB. The Gazelle can be configured with additional top specs, such as an NVMe SSD up to 2TB. Actually, it can accommodate two such drives, so you can theoretically have 4TB of speedy storage. You can opt for either a 15 or 17-inch display, both of which have a 1080p resolution. Regardless of screen size, you get a full keyboard too, meaning it has a number pad on the right. It ships with Ubuntu or Pop!_OS pre-installed and starts at $1099.

An anonymous reader quotes a report from Futurism: During a Tuesday shareholder meeting, Tesla CEO Elon Musk admitted that the electric car company has a design for a submarine car -- and it sounds like something straight out of a 70s James Bond movie. When asked if the company would ever consider such a thing, according to Business Insider, Musk answered: "Funny you should mention that we do have a design for a submarine car like the one from 'The Spy Who Loved Me.'"

In the 1977 spy thriller Musk referred to, a 1976 Lotus Esprit sports car transforms itself into an amphibious submarine and can be seen exploring the bottom of the ocean. Musk loved the scene so much that he reportedly bought a Lotus Esprit in 2013 for just shy of a million dollars. "I was disappointed to learn that it can't actually transform," Musk told shareholders. "What I'm going to do is upgrade [my own Lotus] with a Tesla electric powertrain and try to make it transform for real." He did however admit that the market for an amphibious Tesla "would be small. Small, but enthusiastic."

An anonymous reader quotes a report from CNN: The renewable energy sector had slightly more installed capacity than coal in April, according to a Federal Energy Regulatory Commission report. That means U.S. power plants can produce more energy from clean sources than coal for the first time in history, according to the SUN DAY Campaign, a nonprofit research group supporting sustainable energy. The breakthrough reflects the plunging cost of solar and wind as well as heightened environmental concern about coal. Also in April, the renewable energy sector was projected to have generated more electricity than coal, according to a separate report published by the Institute for Energy Economics and Financial Analysis. That transition was partially driven by seasonal issues.

In an effort to create a "simplified experience," Google said in a blog post today that Google Photos will stop syncing to Drive in July. Digital Trends reports: The change is sure to be controversial. For many, the fact that Photos automatically syncs to Google Drive is a favorite feature, as it allows for much easier organization of photos. Of course, the change will avoid some confusion. According to Google, the change is aimed at helping "prevent accidental deletions of items across products." In other words, it seems like some users were confused about the fact that deleting a copy of a photo in Photos also means that the image is deleted in Drive, and vice versa. The blog post notes that the two services will still work together to an extent. The company announced a new feature called "Upload from Drive," which will allow users to manually select photos and videos to be imported into Photos. Once the items are uploaded, the files won't be connected, so you can delete the file in one without it being removed in the other.

Additionally, Backup and Sync will continue to work on both Windows and Mac, "so if you store your photos locally and want to then sync them to either Google Drive or Google Photos, you'll still be able to do so," reports Digital Trends. Google also notes that existing photos and videos will stay in both Photos and Drive, but the Google Photos folder in Drive will no longer update automatically.

An anonymous reader quotes a report from TechCrunch: Tesla is currently installing its solar roof product in eight states, according to Elon Musk, speaking at the Tesla Annual Shareholder Meeting on Tuesday. The solar roof-tile project has had a relatively long genesis since being unveiled three years ago, in 2016. In addition to having installations run in eight states, Musk said the solar roof product is currently on version three, and that this version is very exciting to him because it offers a chance of being at cost parity with an equivalent entry-level cheap traditional tile, when you include the cost of utilities you'd be saving by generating your own power instead. Regarding timelines for wider rollout of the solar roof products at the costs he anticipates, his own words probably say it best: "I'm sometimes a little optimistic about time frames -- it's time you knew," he joked at the meeting.

An anonymous reader quotes a report from Ars Technica: Geothermal power sources come in many forms, and they're typically much more subtle than steam shooting out of the ground. In reality, geothermal energy could be a big player in our future mix. That is made clear by the U.S. Department of Energy's recently released "GeoVision" report. The report follows similar evaluations of wind, solar, and hydropower energy and leans on information from national labs and other science agencies. It summarizes what we know about the physical resources in the U.S. and also examines the factors that have been limiting geothermal's deployment. Overall, the report shows that we could do a whole lot more with geothermal energy -- both for generating electricity and for heating and cooling -- than we currently do.

There are opportunities to more than double the amount of electricity generated at conventional types of hydrothermal sites, where wells can easily tap into hot water underground. That's economical on the current grid. But the biggest growth potential, according to the report, is in so-called "enhanced geothermal systems." These involve areas where the temperatures are hot but the bedrock lacks enough fractures and pathways for hot water to circulate freely -- or simply lacks the water entirely. Advancing enhanced geothermal techniques alone could produce 45 gigawatts of electricity by 2050. Add in the more conventional plants, and you're at 60 gigawatts -- 26 times more than current geothermal generation. And in a scenario where natural gas prices go up, making geothermal even more competitive, we could double that to 120 gigawatts. That would be fully 16 percent of the total projected 2050 generation in the U.S. The report also estimates that installations of traditional ground-source heat pumps, which circulate fluid through loops in the ground to provide cooling in the summer and heating in the winter, could be increased 14 times over, to 28 million homes by 2050, "covering 23 percent of national residential demand." When factoring in the limitations for how quickly the market could realistically change, the number only goes down to 19 million homes -- still a massive increase.

Meanwhile, district heating systems, where a single, large geothermal installation pipes heat to all the buildings in an area, could be more widely deployed to more than 17,000 locations, covering heating needs for 45 million homes.

Socguy writes: The Uno-X hydrogen station in Sandvika in Baerum exploded on Monday and resulted in two injuries in a nearby non-fuel cell vehicle. The company operating the station has suspended operation at its other locations following the explosion. With the refueling network crippled, Toyota and Hyundai have announced that they are temporarily halting sales of fuel cell vehicles. Jon Andre Lokke, CEO of Nel Hydrogen, the company operating those hydrogen refueling stations, commented: "It is too early to speculate on the cause and what has gone wrong. Our top priority is the safe operation of the stations we have delivered. As a precaution, we have temporarily put ten other stations in standby mode in anticipation of more information."

Here's what Toyota Norway manager Espen Olsen had to say: "We don't know exactly what happened on the Uno-X drive yet, so we don't want to speculate. But we stop the sale until we have learned what has happened, and for practical reasons, since it is not possible to fill fuel now." He added: "This does not change our view of hydrogen, and it is important for us to point out that hydrogen cars are at least as safe as ordinary cars. The hydrogen tanks themselves are so robust that you can shoot them with a gun without knocking them."

MojoKid writes: AMD let loose today with a number of high profile launches at the E3 2019 Expo in Los Angeles, CA. The company disclosed its full Zen 2 Ryzen 3000 series microarchitecture, which AMD claims offers an IPC uplift of 15% generation over generation, thanks to better branch prediction, higher integer throughput, and reduced effective latency to memory. Zen 2 also significantly beefs up floating point throughput with double the FP performance of the previous generation. AMD also announced a 16-core/32-thread variant, dubbed Ryzen 3950X, that drops at $750 -- a full $950 cheaper than a similar spec 16-core Intel Core i9-9960X. On the graphics side, AMD's RDNA architecture in Navi will power the company's new Radeon RX 5700 series, which is said to offer competitive performance to NVIDIA's GeForce RTX 2070 and 2060 series. The Navi-based GPU at the heart of the upcoming Radeon RX 5700 series is manufactured on TSMC's 7nm process node and features GDDR6 memory, along with PCI Express 4.0 interface support. Versus AMD's previous generation GCN (Graphics Core Next) architecture, RDNA delivers more than 50% better performance-per-watt and 25% better overall performance. Greater than 50% of that improvement comes from architecture optimizations according to AMD; the GPU also gets a boost from its 7nm process and frequency gains. Radeon RX 5700 and 5700 XT cards will be available in market on July 7th, along with AMD Ryzen 3000 chips, but pricing hasn't been established yet for the Radeon GPUs.

Long-time Slashddot reader Rei writes: On the face of it, sodium seems like about the worst thing you could make a wire out of — it oxidizes rapidly in air, releases hot hydrogen gas in water, melts at 97.8 degrees Centigrade, and has virtually no tensile strength. Yet, in the late 1960s and early 1970s, the Nacon Corporation did just that — producing thousands of kilometers of high-gauge sodium wiring for electrical utilities — and it worked surprisingly well.

While sodium has three times the (volumetric) resistivity of copper and nearly double that of alumium, its incredibly low density gives it a gravimetric resistivity less than a third of copper and half of alumium. Priced similar to alumium per unit resistivity (and much cheaper than copper), limitless, and with almost no environmental impact apart from its production energy consumption, sodium wiring proved to be much more flexible without the fatigue or installation damage risks of alumium. The polyethylene insulation proved to offer sufficient tensile strength on its own to safely pull the wire through conduits, while matching its thermal expansion coefficient. The wiring proved to have tamer responses to both over-current (no insulation burnoff) and over-voltage (high corona inception voltage) scenarios than alumium as well. Meanwhile, "accidental cutting" tests, such as with a backhoe, showed that such events posed no greater danger than cutting copper or alumium cabling. Reliability results in operation were mixed — while few reliability problems were reported with the cables themselves, the low-voltage variety of Nacon cables appeared to have unreliable end connectors, causing some of the cabling to need to be repaired during 13 years of utility-scale testing.

Ultimately, it was economics, not technical factors, that doomed sodium wiring. Lifecycle costs, at 1970s pricing, showed that using sodium wiring was similar to or slightly more expensive for utilities than using alumium. Without an unambiguous and significant economic case to justify taking on the risks of going larger scale, there was a lack of utility interest, and Nacon ceased production.

"Nuclear experts have warned against re-opening a 43-year-old Scottish nuclear reactor riddled with cracks over fears of a meltdown," writes the Daily Mirror.

An anonymous reader quotes their report: Hunterston B nuclear power plant was shut down last year after it was found that Reactor 3 had almost 400 cracks in it -- exceeding the operational limit. EDF, which own the plant in Ardrossan, Ayrshire, are pushing to return the reactor to service at the end of June and July and want to extend the operational limit of crack allowed from 350 to 700. However, the plans to reopen the plant have sparked fears it could lead to a nuclear meltdown similar to the 1986 Chernoybl disaster.

Experts have warned that in the very worst case the hot graphite core could become exposed to air and ignite leading to radioactive contamination and evacuation of a large area of Scotland's central belt -- including Glasgow and Edinburgh. According to Dr Ian Fairlie, an independent consultant on radioactivity in the environment, and Dr David Toke, Reader in Energy Policy at the University of Aberdeen, the two reactors definitely should not be restarted...

The plant, which is more than 40 years old, can generate enough electricity to power more than 1.7 million homes, and is one of Britain's eight nuclear plants which provide around 20 percent of the country's electricity.
Nuclear expert Professor Neil Hyat reminds The Sun that the reactor will be shut down by 2030 -- and "possibly earlier."

McGruber quotes TechCrunch: Maker Media Inc ceased operations this week and let go of all of its employees — about 22 employees" founder and CEO Dale Dougherty told TechCrunch. "I started this 15 years ago and it's always been a struggle as a business to make this work. Print publishing is not a great business for anybody, but it works . . . barely. Events are hard . . . there was a drop off in corporate sponsorship." Microsoft and Autodesk failed to sponsor this year's flagship Bay Area Maker Faire.

But Dougherty is still desperately trying to resuscitate the company in some capacity, if only to keep MAKE:'s online archive running and continue allowing third-party organizers to license the Maker Faire name to throw affiliated events. Rather than bankruptcy, Maker Media is working through an alternative Assignment for Benefit of Creditors process.

"We're trying to keep the servers running" Dougherty tells me. "I hope to be able to get control of the assets of the company and restart it. We're not necessarily going to do everything we did in the past but I'm committed to keeping the print magazine going and the Maker Faire licensing program." The fate of those hopes will depend on negotiations with banks and financiers over the next few weeks. For now the sites remain online.

A PNAS paper published this week outlines a plan to establish 70 islands of solar panels, each 328 feet in diameter, that sends electricity to a hard-hulled ship that acts as an oceanic factory. "This factory uses desalinization and electrolysis equipment to extract hydrogen gas (H2) and carbon dioxide (CO2) from the surrounding ocean water," reports Ars Technica. "It then uses these products to create methanol, a liquid fuel that can be added into, or substituted for, transportation fuels. Every so often, a ship comes to offload the methanol and take it to a supply center on land." From the report: The researchers estimated that we would need approximately 170,000 of these solar island systems to be able to produce enough green methanol to replace all fossil fuels used in long-haul transportation. While that seems like a lot, it's theoretically possible, even if we restrict these systems to ocean expanses where waves don't reach more than seven feet high and there's enough sunlight to meet the system's yearly average need.

Still, the authors admit that this is just the description of a possible prototype: whether it's practical to build or not will depend on the cost of the technology that supports the system, as well as the cost of competing forms of energy used in transportation. Cleaning and maintaining this equipment in a marine environment is also a concern, and the researchers admit that there may be room for alternate setups (like making another fuel instead of methanol) that might make more economic sense. For now, though, it's a compelling idea to avoid additional fossil fuel extraction that is within reach using existing technology.

In a new paper, researchers at the University of Massachusetts, Amherst, performed a life cycle assessment for training several common large AI models. They found that the process can emit more than 626,000 pounds of carbon dioxide equivalent -- nearly five times the lifetime emissions of the average American car (and that includes manufacture of the car itself). MIT Technology Review reports: The researchers looked at four models in the field that have been responsible for the biggest leaps in performance: the Transformer, ELMo, BERT, and GPT-2. They trained each on a single GPU for up to a day to measure its power draw. They then used the number of training hours listed in the model's original papers to calculate the total energy consumed over the complete training process. That number was converted into pounds of carbon dioxide equivalent based on the average energy mix in the US, which closely matches the energy mix used by Amazon's AWS, the largest cloud services provider.

They found that the computational and environmental costs of training grew proportionally to model size and then exploded when additional tuning steps were used to increase the model's final accuracy. In particular, they found that a tuning process known as neural architecture search, which tries to optimize a model by incrementally tweaking a neural network's design through exhaustive trial and error, had extraordinarily high associated costs for little performance benefit. Without it, the most costly model, BERT, had a carbon footprint of roughly 1,400 pounds of carbon dioxide equivalent, close to a round-trip trans-American flight. What's more, the researchers note that the figures should only be considered as baselines. Using a model they'd produced in a previous paper as a case study, the researchers "found that the process of building and testing a final paper-worthy model required training 4,789 models over a six-month period," the report states. "Converted to CO2 equivalent, it emitted more than 78,000 pounds and is likely representative of typical work in the field."