Paid "research" by a company that would benefit from the result? Check.

Questionable self selection of data sources ("We focus only on conversations in the United States to align with occupation and work activity information from O*NET.", "Note that Copilot-Thumbs may not be representative of overall task success, as some types of users may be more likely to provide feedback, or some types of tasks may be more likely to elicit feedback from users.")? Check.

Oversimplification? Whoo boy, they are killing it on this one. An economist is described as "Compile, analyze, and report data to explain economic phenomena and forecast market trends, applying mathematical models and statistical techniques.", which is simplified to "Forecast economic, political, or social trends.", which is simplified to "Analyze market or industry conditions.", which is simplified to "Analyze market or industry conditions." Check.

Using the thing being studied in the study ("we use a GPT-4o-based LLM classification pipeline to identify all intermediate work activities (IWAs)"? Where's Kramer? Why don't you just tell me the name of the movie you would like to see? Check.

Conclusions that defy basic logic. "Passenger Attendants" scores high on can be done by AI. Flight Attendants who bring you drinks and evacuate you in an emergency are going to be replaced by AI? Or maybe redcaps, the AI is going to carry your bag to the plane or train? Or maybe the sleeper car attendant on the train who makes your bed. Check.

Is AI taking some jobs? Probably. Is it taking most of them? No way. The tech economic bubble burst. Too many tech things have reached saturation, there's no more up and to the right for them. Companies predictably have gone into cost cutting mode as a result, and the stock isn't being juiced by 10x growth. So they are selling an AI snow job to cover up that the core businesses have reached saturation and aren't going to generate 10x results anymore. Tech is in for a hard time in the near future, not because of AI, but because people have as much social media, streaming, etc as they want, and everyone now has it. In fact many are fed up and giving up parts of it, because they have learned it's garbage. There's no new people to feed into most of the systems anymore.

That's right, people need to stick to reliable, high quality sites like /.

I think youâ(TM)ve missed how much money AI companies are lighting on fire on both compute and being first. If it cost a billion in compute to solve the captchas but they got the data, they would do it. The fundamental problem is companies are being rewarded with billions of investment and stock growth for this behavior. They will continue until the money dries up.

This has little to do with AI and and everything to do with Section 174(c)(3) in the tax law. https://www.corumgroup.com/ins... First, this caused companies to hire less developers, in particular startups that couldnâ(TM)t take the tax bill. Second it caused women companies to reclassify people as other roles since calling them software developers was no longer a write off. The tax law is 90% of the jobs lost, AI is 10% at best.

The root of the answer is economic, The Productivity Pay Gap.

People are always for their own self preservation first. If you want them to help others, be tolerant of others, or generally work on making a better society they have to feel like their own self preservation is not being threatened. Over time the disconnect in that graph has grown, leading to more and more people worried about their own preservation. Unable to afford housing, food, transportation, they start to worry only about themselves and start to get mad when "those people" get some leg up that they don't get.

I think the major difference is long distance lines. Today someone building a data center might need to fund a mile or two of high voltage distribution to the nearest trunk line or substation. I'm told this costs between $3-$8M/mile. When building a mega-data center an extra $10M to get power there is a rounding error. It's also important to note that 12-15% of the power is lost over long distances. When moving 2GW

that's 240MW of lost power! Someone has to pay for that as well.

What's happening though is that transmission into entire regions is being used up. This will require new transmission lines over longer distances. Let's assume the low end of the cost (much of the lines are in rural areas), and let's assume it takes 3 new major trunk lines to increase capacity in a region. Further, let's assume 150 miles per line. That's 450 miles at $3M/mile. Adding $1.35B in cost to new data centers is NOT a rounding error. And worse, this is going to happen in multiple regions in approximately the same timeframes.

Also, let's not forget the land area and other resources. A 200 foot wide transmission corridor going 150 miles requires 3,600 acres of land. So those three new trunks are around 10,000 acres. SMR's might need only 50-100 acres.

Data centers and power plants have to be close together. I would prefer to site data centers near existing power plants away from major metros. But I get why people are looking the other way, moving the power plant to the data center. There's no way transmission over long distances can be cost effective.

Your high level analysis of the Vogtle debacle is approximately correct. However, I do not think that is what the data center industry is talking about when they reference nuclear power.

SMR aka Small Modular Reactors are the talk of the town in data center circles today. An example would be the new units from NuScale Power. Theoretically these can be placed in a 3-5 year timeline, maybe faster if production was ramped up.

The general idea is to have a number of 50-300MW reactors on-site. There's big wins on the data center side. Costly and often held up by NIMBYs high tension lines over long distances are no longer required. Most of the battery and generator backup systems are no longer needed, simply run N+2 or N+3 SMRs. There are other designs, including a lot of buzz around Molten Salt Reactors, hybrid plants with batteries and solar, and then more far out ideas. Some are really close to commercial ready, while some are quite far away.

I have very mixed feelings on this concept. While I understand the engineering enough that I know it's not another 3 mile island or Chernobyl, it still means creating, transporting, and disposing of nuclear fuel. I do like the idea of some mega-data center sites further away from population centers. If these massive grey boxes are going to be built with on-site power, put them away from people. Further, they could be sited to take advantage of existing infrastructure. There are plant sites, like old aluminum factories, that consumed gigawatts of power. The transmission lines already go to them. There are existing power plants all over that could be upgraded and modernized to power a nearby data center campus.

What I do know is a lot of people are throwing a lot of money at AI right now. While I'm not sure it's a good investment, give a company enough billions and a lot of things that people would never guess could happen, happen. Are the goals ambitious? Absolutely. Are they impossible? I'm not so sure.

Code that makes money is not available for public consumption and thus not ranked. This list is at best the most popular languages for open source.

At many of the data centers and businesses around here the shredder truck shows up on site. Company personnel scan serials and put them in. A supervisor watches. Whole drives are not allowed out of the parking lot, and arenâ(TM)t allowed out of the cloud providers hands. No one else touches them.

The sad thing is, you likely have no idea how inefficient, insecure, and bug riddled that code is because you do not have a proper programming background. If this becomes popular, people like you will triple computings carbon footprint, provide lots of new ways for things to be hacked, and introduce subtle errors into all sorts of data sets that will slowly multiply as additional poorly writte. Code processes them. I have a real fear that over the next 10 years computing will seriously regress in quality.

I know a couple of (very minor) internet âoecelebritiesâ. That is they have blogs and such that get hundreds of thousands of views regularly. They tried ChatGPT to edit posts. Universally the ChatGPT posts had 1/3 to 1/4 the views of their own writing. We are not sure, but we theorize people like them because of some unique style or quirk, and ChatGPT goes for a bond algorithmic average. I wonder if after switching to ChatGPT editors if your publications will find declining readership for the same reasons.

The longest delivery route in the entire country is.......Longest Rural Delivery Route

Clarinda, IA, 181.4 miles daily.

But that is unusual, this news article reports.......USPS Could Serve Nearly All Its Mail Routes in Electric Vehicles

The average postal route requires 24 miles of driving and nearly all of them are less than 70 miles.

That E-Transit provides 5x the range for the average route, and almost 2x the range of the "nearly all" figure. The article suggests 99% of the routes fit within electric vehicle range with current technology.

Remote work requires good broadband. West Virginia ranks in the bottom 10 by most measures. A cozy cabin in the woods is no good for remote work if all it has is dial up.

Remote work is also about the amenities. People live in cheaper places so they can afford to go to concerts, ball games, enroll their kid in camp, etc. West Virginia is way short on these, and the few people who move there will notice it fast.

Many motherboards advertise "Hardware RAID", but are in fact what we call "fake raid". They have some hardware acceleration features on the motherboard, but the heavy lifting of the RAID is done in the driver. Some of these are Windows only as a result, while some are supported by various Open Source drivers now. See ABMX Servers for an article on the differences.

If you use RAID, hardware, software, or fake, you also need to consider the sort of drives. Drive firmware is different for RAID arrays than for single drive applications. The most important difference is TLER/ERC drives will retry a bad read/write MUCH fewer times before erroring out. If you use firmware configured for many retries in a single drive application it can absolutely destroy the performance of your RAID. Rather than the RAID being able to move on to the other drive(s) and/or remap the sector that failing drive just hangs the whole thing with retries. For a while this was a user configurable parameter in many drives with SMART, but most manufacturers have now limited it to RAID capable drives.

Why are "RAID" drives more money? Well, there are several reasons, but one of the big ones is vibration. When you have multiple drives in a chassis doing synchronized tasks they can end up vibrating each other into poor performance. A rather famous video of a guy screaming into a RAID array proves the point. Non-RAID drives often omit some of the vibration dampening features and it leads to worse performance and premature failure, particular when using 5+ drives in a single chassis. Obviously this does not apply with modern SSD's, another case where SSD's are superior.

Generally my advise would be to use a pair of disks in a RAID1 mirror with a software driver for an end user machine, like a desktop used to play games. In a server application where multiple drives are required for capacity I'd recommend a dedicated hardware RAID card from a quality vendor driving RAID spec hard drives. YMMV, plenty of folks get away with other configurations.

The iPhone will happily connect to directory services, like LDAP. My work phone has all of my work contacts automatically, because my business sets that up right.

The iPhone will AirDrop a contact to another iPhone. When you meet someone they can drop their contact info to you, just like handing you a business card, no typing anything in on your own.

The iPhone can share a card via e-mail or MMS, both import and export, using VCARD. Someone can text you their card, even from an Android phone, and it will happily import into contacts just fine.

The iPhone will share contact lists via iCloud with all your other Apple devices. For the ones that do need manual entry, edit on your Mac, or via their iCloud.com web interface. The lists can also be exported and shared as VCARD lists with anyone, anywhere.

The App Store has a dozen lightweight apps that will take a photo of a business card and create a contact entry for you based on OCR.

Apparently Marissa knows none of these options, and so she's manually typing them when she meets someone. That's not Apple didn't provide the functionality, that's Marissa didn't do basic investigation on what the product can do before launching off on a whole new company to duplicate existing functionality from Apple and others. No wonder Yahoo failed. Did none of the yes-men around her not point these things out when she had this crazy idea?