I don't know if it's quite that. Theranos had experts saying "this doesn't seem physically possible" and never really demonstrated a working product. Classic smoke and mirrors job. I use Anthropics models routinely in my job now, which even a year and a half ago I would have been pretty skeptical of. They're pretty goddamned good. Not perfect, not error free, not a brain replacement. But good enough that even for pretty complex, domain specific tasks I'm spending more time in "review mode" than "doing mode" for most things anymore. I've basically had to admit I was wrong on the pace and capabilities of these things... I thought these could be a "real" disruption in a decade or two, but they're here now. The trick has been trying to figure out how to navigate using them, the capabilities have changed so dramatically so fast we're struggling to lay plans, estimate needed headcount, etc.

There is just a huge amount of white collar busywork that took significant time but was just different enough from iteration to iteration and just complex enough to automate that it never made sense to do it. Sometimes you could plug an intern or a junior into doing that work but not uncommonly the deadlines were just too short or the complexities just a little too high to rely on that. From my experience Opus/Fable are able to fill that hole by and large now, and it is kind of spooky. In my domain probably the biggest hesitation we still have is navigating attorney-client privilege and to what degree the use of those models creates legal issues that didn't used to exist, but thus far the time savings has been so extreme the dictate has been to keep using them.

I think the more realistic take isn't that this is smoke and mirrors a la Theranos, but that these AI firms are massively overextended and they know it. But because they're in an arms race they can't stabilize. A slowdown would let them shore up their financial backing, grow headcount in a coherent way, firm their products up, establish new revenue streams, and push back on the current media narrative that AI is the end of civilization. All huge positives for them, even if they genuinely also believe the technology to be dangerous and want to "do the right thing". And frankly I think they realize the big money here will be selling training and model customization (every corporation has its own model, likely multiple for different subdomains), but that's going to need a bunch of these monster datacenters to actually finish building coupled with a pretty massive corporate push to improve data that currently isn't in a form that's useful to AI training. That sets the timeline for the big profits at 5-10 years from now. At their current rate of spend most of these AI firms will struggle to survive that long, and that's not accounting for the risk of populist backlash to the technology.

The issue isnt the components, those are simple. The issue is a huge amount of this equipment is bespoke, and the facilities and expertise are gone in the US to make them (GE has a facility theyve tried to rapidly grow, but itll be nowhere near enough for quite a while). With the explosion in datacenter growth and accompanying power infrastructure the lead time on equipment is five to ten years. Theres alternatives being developed domestically that are mass-producable that use some clever new techniques to remove the bespoke aspects of the equipment, but it'll be probably another 10 years before that tech is proved out in grid trials and starts scaling.

As someone who's job is deep in the thick of grid planning, id say the concern here is legitimate but the ship has sailed. This is one of the many things that never should have been allowed to be offshored, but since it was the realities on the ground just dont support an executive order in this form. The kind of targeted, government subsidized intervention Biden did with chips and transformers and all kinds of other infrastructure is the smart way to handle these issues. This is the dumb way. We should give the half dozen companies working on the next generation of this technology government funding to accelerate, while sucking up as much foreign production of the current generation as we can. What trump is doing is instantly kneecapping growth while aiming our sights backwards instead of forwards.

Coal looked like it was on the way to be phased out sometime mid-2030s, but now I'd say it's roughly stabilized. You'll still see some forced retirements in areas trying to me more aggressive about clean energy, and it's not impossible that the supply chain around coal could collapse entirely and take the rest of the coal fleet with it given the right circumstances. But the demand is there, and utilities are backing off plans to phase coal out. A large contributor to that has been gas pipeline opposition, which has ironically served to prop up the coal industry rather than allowing its replacement with cheaper and just overall more environmentally friendly gas plants that neatly slot in for coal.

One thing articles like this conveniently leave out is in the spring and fall demand is much lower (and there's much less reliability risk), and as such these large thermal units that need maintenance are scheduled to be taken out of service for preventative repairs in these periods. So the monthly percentage becomes very misleading. Annual relative output averaged over a handful of years to reduce weather anomalies is a much more honest progress baseline.

Exactly. Social media is an ecosystem, this is degrading the ecosystem for young users. It could plateau, in which case more intervention is warranted. But more than likely, just like in a real ecosystem, it begins to degrade things fundamentally. Fewer kids get access young, so the social pressure to be online subsides. Parents now have firm backing not to allow social media use, and feel increasingly empowered to push back on child access. With fewer users in their peer group engaging and creating content the addictive qualities diminish and more kids start to opt out intentionally. It's not about eliminating every user, in the same way age verification for alcohol isn't about stopping every kid from getting drunk. It's about degrading it systemically.

Right now, we have a generation of addicts. Addicts will do what it takes to get their fix, and many parents will enable them because that's easier than dealing with an addict in withdrawal. But that will wane over time. I had an interesting conversation with some other parents along these lines, where they were convinced they'd need to get their preteen a smartphone for social reasons, even knowing it would be detrimental. And the obvious argument against that would be "if it was technically legal to give your kid meth, would you do it if 95% of other children's parents were doing it?" And they were still very on the fence about any kind of restriction. Social pressure is extremely powerful, and right now the social pressure to give children very addictive stuff is strong.

$20,000 of gas avoided in three years? Holy shit. In fairness, that's a wild outlier of a use case. I pretty seriously looked at an Ioniq a couple years ago, liked the car, but just couldn't get anywhere near cost parity with a gas vehicle even with a moderate use daily driver. We do about the "normal" US amount of driving, 10 thousand miles a year give or take, and the cost difference even with the EV subsidy was just way too high. I mean just back of the envelope, if you figure 24mpg fuel economy for the gas car and 10,000 miles driven, at $3 per gallon that's only $1250 of savings a year in fuel. The upfront cost was easily in the tens of thousands of dollars more range for EV versus a similarly spec'd out gas car, and because of that higher cost to buy my yearly insurance cost increase was going to offset most of the EV fuel savings.

Even if the cost of gas goes to $10 a gallon in the US with this war, it would need to stay there for many years to justify the cost gap between a new EV and a new gas car with comparable features, for an average driver. If you're putting on 40K miles a year as an uber driver or something that math is a lot more favorable though.

Averages aren't really a useful way to determine if a generating asset will have an impact in wholesale markets. For one, wind output is highly variable but follows broad seasonal and daily patterns. It's very possible for the bulk of the output of the wind farm to be in periods when power would already be cheap. One particular bit of trouble you run into is the wind lowers the price in off-peak hours, but the system still needs expensive power units to stay online to help cover the peak hours (imagine a coal plant, for example). The market then ends up setting the price higher in peak hours to incentivize those units to stay on, and it's not uncommon for large thermal units to run at a substantial loss when wind is available but make that money up in the peak hours. They'll throttle their output to minimize this, but there's some baseline level those plants have to output at just for mechanical reasons.

In broad terms, what that means is "cheap" wind or solar power can, depending on the local situation, make the cost of electricity go up even though their marginal cost of production might be lower on average. Which feels counterintuitive, but in a roundabout way it's the cost of reliability being baked into the wholesale price and wind/solar have an inherent reliability penalty. If all you cared about was cost then you're almost certainly going to just run the baseload power units that you need for reliability more often and forego the wind/solar capital expense. This dynamic is the whole reason all these subsidy schemes for wind and solar exist... the raw financial incentive for wind and solar isn't there in the wholesale market, and is particularly bad when you're comparing new-build wind and solar versus an existing fossil or nuclear power plant (since they're already a sunk cost). If the goal is to reduce CO2 then you need those existing facilities to throttle down or close, but using them is almost certainly cheaper than building a brand new asset. So the new asset needs incentives.

Now if your renewable production usually lines up with your demand pretty well, like in the case of homes that overchill during peak solar hours and use that "thermal bank" during peak consumption hours (usually at dusk, when solar output drops off) then this is substantially mitigated. And to some degree, with the right pricing schemes, you can shift residential and industrial demand partially into the hours when wholesale power is cheaper. But by and large if you have wind and solar at any scale then it's going to take significant amounts of batteries or massive transmission ties into other regions with low wind/solar penetration for that power to be regularly useful, which are themselves expensive and raise the cost of wholesale power.

What I dont understand is who is this for? In buisness contexts I get it, but personal use? There is the very occasional project where, if I could trust something like this, it might save me marginal amounts of time. But the vast majority of the time this would be useless. Where's the market? Can anyone here explain exactly what some massive agent swarm is supposed to do on my computer other than be a massive liability? It seems to me almost like theyre looking at time spent on computers like its a time suck that can be eliminated, but the vast majority of the time people are doing it for enjoyment anyway. Like, if I want to buy a watch, I want to research and understand that myself. I dont want to pay an AI provider to shit out a watch reccomendataion while I mow the lawn.

Texas has had the most ICE arrests under many presidents because its an easy place for immigrants to both cross the border and surrender, and many border crossers just dont go further. The high statistic is just border apprehensions. There arent masked ICE agents in Texas urban areas en mass running cars off the road, teargassing, beating, and murdering people. The comparison is bullshit.

Comically Trumps goons accused Minnesota law enforcement of not cooperating with ICE detainers and the local law enforcement provided receipts that it just wasn't true. The "non cooperation" is just made up. People there are up in arms about the blatant constitutional violations and harassment. The cops were openly supportive of ICE up until the murders, and subsequent harassment by ICE of off duty MN cops.

Minnesota has one of the smallest illegal immigrant populations in the nation. Do you know how badly you need to want to be here to leave someplace tropical and end up in Minnesota? There is literally no mathematical or logical justification for what is happening, other than an attempt to make an example of a political opponent. Its intimidation by a thug, nothing more. Glad to see Minnesota give those assholes a rash.

Its interesting to think about what an anomaly the last hundred or so years has been. Prior to easy photography and radio/telegraph virtually all information would have been secondhand and easily manipulated or misconstrued. At the founding of the US there would have been many printers in any major city, each with agendas. There were prominent "debates" held through these publications that were done by pseudonym and were often false or salacious. Realistically thats part of what lit the fire of the American revolutionary war. When you think about it, it all looks a whole lot like the modern partisan media landscape with all its manipulations.

I dont know how this all unfolds, and its entirely possible modern technology amplifies the bad aspects of this way more than in the past (its a lot easier to be fooled by a realistic image than a political cartoon, for example). But i do take a little comfort in the fact that we've been here before.

China might have more water logistics than the US but remember a ridiculous volume of goods travels through the Mississippi and great lakes. I'd guess the two nations are within earshot of each other there too.

Its interesting, for as much rightful shit as the US takes for not having more developed commuter rail it has a substantial freight rail network. In fact theres a lot of youtube videos arguing that fundamentally the rail differences between the US and Europe in particular are largely prioritization of people vs freight rather than lack of investment.

*For now*

The purpose of dumping like this is to disintegrate competition, as it's not physically possible to keep up with a competitor who can sell at a loss indefinitely via funding from the state or other unrelated business assets. It's part of why monopolies are supposed to be illegal, and what the Chinese have been doing violates international trade rules (which are laughable in their enforcement). The Chinese have been pulling crap like this for two decades, with the explicit aim of making western nations too dependent on their industrial base to take any serious action against them on other issues (Taiwan annexation, human rights abuses, etc.).

But the end state of these kinds of tactics is those Chinese firms will be all that remains, and free to jack up prices. Because of the cost and complexities involved in entering the industry it's unlikely competition would re-emerge globally, and if it did the Chinese government could just temporarily funnel money back into the industry to destroy foreign competition again. Even just the threat of that would be enough to kill most private investment in the sector outside China.

People need to reckon with the fact that cheap Chinese goods do not exist because the Chinese are more efficient, frankly what efficiency advantage they have in most industries is small. Those cheap goods are enabled by a mix of environmental damage, human damage, and state funded economic warfare. Those are costs, even if they're hidden from the consumer. It's incumbent on governments around the globe to account for them honestly, and restrict trade to compensate.

Yeah, the US model is well known to be worse than the European models overall, but sometimes better in select scenarios. Typically I hear forecasters mixing predictions from the various models based on circumstance.

But even ignoring that... I'd expect that a trained model like Googles would dominate a relatively tepid hurricane season. What will be interesting is if it does a better job on an extremely atypical, high impact (landfall) storm system, and can replicate that repeatedly. In general in modeling (anything, not just weather) there is no one "right" model, as tempting as it is to think so. Various models make tradeoffs based on the parameter of interest, the circumstance, etc. I think there's a lot of potential for newer AI/machine learning techniques to improve modeling across a range of disciplines, but there is no "one ring" of models. It's just not a thing.

You're assuming the next step here isn't punitive changes in tax or monopoly law to punish offshoring. All the executive has to do is threaten convincingly that it's in these tech companies best interest to bring jobs back to the US, or they'll find themselves on the wrong end of a whole lot of executive action. Whether it's from Trump or the next populist, the threat of having your tech monopoly broken up would be existential. Until recently the large corporate players thought they were invincible, but Trump has shown that it's really just norms and popular sentiment that enabled them to gut the US workforce in favor of cheap overseas labor. Those norms are shattered, and popular sentiment in the US was turning against globalism even before Trump.

If I were in charge of these corporations I'd read the writing on the wall and begin onshoring with US native workers as a hedge. Tech giants have gotten too comfortable with the idea they can outmaneuver any consequences of US policy with lawyers and clever logistics, but the reality is they're extremely vulnerable.

That's not how electricity prices work.

In non-wholesale market areas the cost of electricity is set by the government. The net return to the utility is fixed and pretty low, low to mid single digits percent. Monthly or yearly price variation in those regions are virtually always related to pass-through costs like fuel or storm damage which the utility "passes through" without markup to the customer. New large loads, like a datacenter, basically get explicitly approved by the state via regulatory mechanisms to ensure they don't raise prices for consumers in excess of the benefits they provide the region.

In wholesale markets prices are set by the market itself (which an independent regional nonprofit runs), with generators bidding into the market with cost curves. That works a little differently market-to-market, but basically the utility calculates their breakeven cost for each generator and submits that. The market stack-ranks all the generators bid in, in real time (say every 5 minutes). The controllers for the grid then send orders to the generators whether they should be online, and at what generation level. There's all kind of complexities with that I won't get into. The key observation there however should be "wait, if they're bidding in at cost how do they make money?" And the answer is the unit "on the margin", i.e. the unit that is the most expensive that is required not to have blackouts, sets the market price in any given 5 minute period. That "most expensive" unit nets basically no money. Everybody else makes a profit of whatever the difference is between the most expensive unit's generating cost and their generating cost. There are independent market monitors that make sure no generator/utility are exerting market influence to game the system (which is illegal).

The author of the blog post is picking up on something real, which is that transmission constraints are increasingly impacting wholesale prices. There's quite a bit more going on than that though, and more transmission won't solve all of it.

One issue is that the price in a given hour is increasingly set by a marginal generator that is totally divorced in price from renewables. In a perfect world those higher cost marginal generators would eventually be replaced by the lower cost technology, but in practice the lower cost technology (solar, wind) is incapable of reliably serving electricity. So the more expensive generators stick around because they have to, and the renewables operators just increasingly pocket the difference. Most grid operators use a separate market that clears once a year and compensates reliability to ensure the grid doesn't collapse because stable generators retire. That has its own insufficiencies, but the end result is that the wholesale price of electricity is increasingly meaningless for sending price signals to generators as to whether they should stay operational. It really only works for telling a generator if they should be outputting in any given minute, not if a plant should continue to operate.

Another issue is that because of the way intermittent generation (wind, solar) warp the market the wholesale cost of electricity increasingly means nothing to consumer bills either. Take California, with the ironically stable (if high) wholesale prices but the soaring retail costs. Why is that? It's because incentives are warping the generation market, and those incentives are funded outside of wholesale electricity cost. Home solar rebate programs. Transmission infrastructure cost (solar/wind are typically far from power consumers and require a lot more/bigger power lines). Insurance cost (more transmission infrastructure for solar/wind increases the likelihood of things like wildfire, which increases insurance cost). Cost to build battery storage. Etc. These things all end up as line items in a homeowners bill, and they're functionally direct costs of wind and solar.

My personal take, as someone who has worked many years in the electric utility industry, is that the wholesale markets (MISO, PJM, etc.) are intrinsically broken. Price signals are not robust enough nor forward looking enough to accurately steer decision making. A more wholistic, top down decision making process just simply works better, and allows better control over things like carbon targets (especially when you have a range of goals among participating states). These market constructs frankly weren't devised very well from the beginning, and the problems with them have grown exponentially as wind/solar/other technologies have appeared and become cost-viable. The alternative is what we had before, and what 1/3 of the US still has: vertically integrated utilities, either privately owned and heavily regulated or outright owned by a government entity.

Right, because the treatment for addiction to heroine is to make the world around you so fantastic you dont feel like doing heroine anymore. Totally works. Honest.

At this point, given how terrible social media and mobile games have been found to be for childhood development, id almost rather my kid got addicted to a chemical than a phone. At least that has practical solutions. Very difficult to remove kids from a digitally saturated society. Giving a pre-teen a smart phone should be flat out illegal.