US AI Action Plan

The US has long since exhausted it's "easy" oil/gas reserves. Yes, there's tons more down there, but it's increasingly hard to get to. Lots of extraction methods only make sense when the price for oil is above some amount.

If the rest of the world standardizes on solar+battery, demand for oil goes down, and so will the price. Which in turn makes US-produced oil not cost effective to extract, and domestic energy production collapses in favor of cheap foreign imports.

And then we're worse off in several different ways.

Ice caps? Try human beings.

Increased Mortality: Projections indicate an additional 14.5 million deaths by 2050 due to climate-related impacts like floods, droughts, heatwaves, and climate-sensitive diseases (e.g., malaria and dengue).

Economic Losses: Global economic losses are predicted to reach $12.5 trillion by 2050, with an additional $1.1 trillion burden on healthcare systems due to climate-induced impacts. One study estimates that climate change will cost the global economy $38 trillion a year within the next 25 years.

Displacement and Migration: Over 200 million people may be displaced by climate change by 2050, with an estimated 21.5 million displaced annually since 2008 by weather-related events. In a worst-case scenario, the World Bank suggests this figure could reach 216 million people moving internally due to water scarcity and threats to agricultural livelihoods. Some researchers predict that 1.2 billion people could be displaced by 2050 in the worst-case scenario due to natural disasters and other ecological threats.

Food and Water Insecurity: Climate change exacerbates food and water insecurity, leading to malnutrition and increased disease burden, especially in vulnerable populations. For example, a significant increase in drought in certain regions could cause 3.2 million deaths from malnutrition by 2050. An estimated 183 million additional people could go hungry by 2050, even if warming is held below 1.6°C.

Mental Health Impacts: Climate change contributes to mental health issues like anxiety, depression, and PTSD, particularly in vulnerable populations and those experiencing climate disasters or chronic changes like drought. Extreme heat has been linked to increased aggression and suicide risk. Studies also indicate that children born today will experience a significantly higher number of climate extremes than previous generations, potentially impacting their mental well-being and sense of future security.

Inequality and Vulnerability: Climate change disproportionately affects vulnerable populations, including low-income individuals, people of color, outdoor workers, and those with existing health conditions, worsening existing health inequities and hindering poverty reduction efforts.

I specifically refer to the question of who will own the IP and economic might to lead in the clean energy market. Who will innovate? Who will build industrial capacity and know how, etc. It seems we’ve ceded the field

Not just strict energy production. Especially when it comes from sources of energy increasingly infeasible and unpopular.

Nuclear is clean, but has other drawbacks. "Solar+Storage is so much farther along than you think": https://www.volts.wtf/p/solarstorage-is-so-much-farther-alon...

Everything I've read recently has emphasized that new nuclear installations will have difficulty competing with solar and storage.

Having a non-emitting form of base load is important, and nuclear has a place there, but it many applications it's just not cost competitive with renewables.

Nuclear fission is more expensive per kilowatt than solar and forces you to go through a lot more trouble to contain risk.

Maybe if fusion was viable, that'll change, but until then nuclear just doesn't make any sense.

I really don't understand HN's love affair with nuclear.

Uranium mining produces significant toxic waste (tailings and raffinates). Fuel processing produces toxic waste, typically UF6. There is some processing of UF6 to UF4 but that doesn't solve the problem and it's not economic anyway. Fuel usage produces even more waste that typically needs to be actively cooled for years or decades before it can be forgotten about in a cave (as nuclear advocates argue).

And then who is going to operate the plant? This administration in particular is pushing for further nuclear deregulation, which is terrifying. You want to see what happens without regulation? Elon Musk's gas turbines in South Memphis with no Clean Air permits that are spewing pollution [1].

That's terrifying because the failure modes for a single nuclear incident are orders of magnitude worse than any other form of power plant. The cleanup from Fukushima requires technologies that don't exist yet, will take decades or centuries and will likely cost ~$1 trillion once its over, if it ever is [2].

And who's going to pay for that? It's not going to be the private operator. In fact, in the US there's laws that limit liability for nuclear accidents. The industry's self-insurance fund would be exhausted many times over by a single Fukushima incident.

And then we get to the hand waving about Chernobyl, Fukushima and Three Mise Island. "Those are old designs", "the new designs are immune to catastrophic failure" or, my favorite, "Chernobyl was because of mismanagement in the USSR" like there wouldn't be corner-cutting by any private operator in the US.

And let's just gloss over the fact that we've built fewer than 700 nuclear power plants, yet had 3 major incidents, 2 of them (Chernobyl and Fukushima) have had massive negative impacts. The Chernobyl absolute exclusion zone is still 1000 square miles. But anything negative is an outlier that should be ignored, apparently.

And then we get to the impact of carbon emissions in climate change but now we're comparing the entire fossil fuel power industry vs one nuclear plant. It's also a false dichotomy. The future is hydro and solar.

and then we get to the massive boondoggle of nuclear fusion, which I'm not convinced will ever be commercially viable. Energy loss and container destruction from fast neutrons is a fundamental problem that stars don't have because they have gravity and are incredibly large.

I have no idea where this blind faith in nuclear comes from.

[1]: https://www.politico.com/news/2025/05/06/elon-musk-xai-memph...

[2]: https://cleantechnica.com/2019/04/16/fukushimas-final-costs-...

You obviously have no idea how much destruction it causes to the environment to get the uranium out of the earth. Maybe educate yourself before putting such nonsense into the world.

Nuclear takes 20 years to build and plants cost $10B.

Rooftop solar starts paying back instantly and can be deployed in $20k tranches. It also requires no additional grid infrastructure and decreases demand on non generating grid infrastructure.

Pretty sure it’s rooftop solar that wins the future.

A certain group of people keep saying that. But that particular idea of "clean" nuclear does not price in the 10.000 years of safe storage of nuclear waste materials (for the most dangerous HLW materials this number can go up to 100.000 years). Do you and your 3500 generations of ancestors volunteer to do this? Then it is cheap and clean. Otherwise it is yet another instance of "privatize the gains and socialize the externalities".

(And let's ignore the fact that humanity barely managed to organize anything that held even a mere 1000 years)

I believe that we should separate the general case of AI from the particular case of LLMs. AI models have been accelerating science for decades, and new technology helps drive economic growth. I am convinced that LLMs are not worth the money we've invested in them but I do believe that more "traditional" AI is a net positive in research fields. Traditional AI has also had a net negative effect on the quality of content from internet publishers (i.e Facebook), but has made it more productive by allowing them to squeeze more blood from the stone.

I think if you don't include LLMs, AI has obviously created economic growth. If you do include LLMs I think the conversation is more nuanced and obviously driven by the same kind of hype that led people to believe that Cryptocurrency is the future of the stock market

We're clearly in an era where the US Govt simply doesn't have enough money to throw at everything it wants to encourage, and needs to develop alternate means of incentivizing (or de-disincentivizing) those things. Sensible minimal regulation is one, there may be others. Time to get creative and resourceful.

Even if they did provide more money, it doesn't mean it'll go to the right place. Government money is not the solution here. Money is already being spent.

It doesn't say anything about open training corpus of data.

The USA supposedly have the most data in the world. Companies cannot (in theory) train on integrated sets of information. USA and China to some extent, can train on large amounts of information that is not public. USA in particular has been known for keeping a vast repository of metadata (data about data) about all sorts of things. This data is very refined and organized (PRISM, etc).

This allows training for purposes that might not be obvious when observing the open weights or the source of the inference engine.

It is a double-edged sword though. If anyone is able to identify such non-obvious training inserts and extract information about them or prove they were maliciously placed, it could backfire tremendously.

That's exactly what the goal is: That everyone uses American models over Chinese models that will "promote democratic values".

The idea is to dominate AI in the same way that China dominates manufacturing. Even if things are open source that creates a major dependency, especially when the secret sauce is the training content - which is irreversibly hashed away into the weights.

They see people using DeepSeek open weights and are like "huh, that could encode the model creators' values in everything they do".

Can a model make so sophisticated propaganda or manipulation that most won’t notice it?

Elon gives an unvarnished look at what they mean by 'unbiased' with respect to models. It's rewriting the training material or adding tool use (searching for Musk's tweets about topics before deciding it's output) to twist the output into ideological alignment.

"unbiased", in the world of realpolitik, means "biased in a manner to further my agenda and not yours".

"major security risks" as in Terminator style robot overlords, or (to me more likely) they enable people to develop exploits more easily? Anyway I fail to see how it makes much difference if the models are open or closed, since the barrier to entry to creating new models is not that large (as in, any competent large company or nation state can do it easily), and even if they were all closed source, anyone who has the weights can run up as many copies as they want.

It's certainly more encouraging than the tone from a few months/ years ago, when there was talk of outright banning open source/ weigh foundational weight models

IMHO it's not empty talk; a lot of the elements of the plan reinforce each other. For example, it's pretty clear that state initiatives that were aiming to place regulatory thresholds like the 10^26 FLOPS limit in Calfornia's SB1047 are going to be targets under this plan, and US diplomatic participation in initiatives like the Council of Europe AI treaty are now on the chopping block. There are obviously competing perspectives emerging globally on regulation of AI, and this plan quite clearly aims to foster one particular side. It doesn't appear to be hot air.

For open source/open weight models it's particularly important because until now there wasn't a government-level strong voice countering people like Geoff Hinton's call to ban open source/open weight AI, like he articulates here: https://thelogic.co/news/ai-cant-be-slowed-down-hinton-says-...

Current AIs are anything but politically neutral.

[dead]

Technically most sources of available energy on or near the planet are the output of fusion in some way, so this tracks.

Your implication is misguided, solar is in fact the most practical way to add more electricity for most countries.

The US generated an additional 64Twh of solar in 2024 compared to 2023. To get the same amount from nuclear you would need to build 5 large reactors in one year.

As for land mass, we can re-use already spent land mass, like rooftops, parking lots, grazing farmland and such. Solar can also be placed on lakes.

So for the foreseeable future there is no actual need for new land to be dedicated to solar.

Even though I think solar is impractical as a primary source for various reasons, it doesn't take a lot.

David MacKay in "Sustainable Energy: Without the Hot Air" did a calculation circa 2010. To fulfill the world's energy needs back then, a 10 km^2 area in the Sahara desert would be sufficient. Even if you scaled that to 100 km^2, it's absolutely tiny on a global scale, and panels have only become more efficient since then.

The challenge of course is storage and distribution, but yeah, in terms of land area, it's not much.

It's worth considering total lifecycle use of water (mining, production and operation) for nuclear and solar.

Solar: ~300-800 L/MWh [0]

Nuclear: ~3000 L/MWh [1]

0: https://iea-pvps.org/wp-content/uploads/2020/01/Water_Footpr...

1: https://www-pub.iaea.org/MTCD/Publications/PDF/P1569_web.pdf

3,000-4,000 acres per GW of production capacity in the US Southwest. According to AI :)

Considering how little use there is for most of that land anyways, it seems like a good option to me.

Also AI training seems like the perfect fit for solar. Run it when the sun is shining. Inference is significantly less power hungry, so it can run base load 24/7.

>How much land mass would need to be covered by solar panels to power this future AI infrastructure

Probably zero agricultural if you mandate all rooftops to be solar. And all parking lots to be covered with solar roofs.

Tariffs, regulatory quick sand, political pressure...

Nothing other than the fact that renewables won't be able to keep up if the AI demand keeps growing the way it has been

They just buy a contract from a power distribution company. They don't care where it comes from.

If you want the PD companies to have a different blend, then they need carrots and sticks.

Demand is too high, same goes for nuclear which takes too long to build.

it needs to be better connected over larger distances i guess. Some 'sunny' countries around the equator are working on it. laying gridlines to other less sunny places and trying to offer solar to reduce carbon taxes or whatever.

i know Saudi, Morocco and China are all massively dumping panels into their deserts, likely more places too. these are great places to put them as it has less impact on environment (less wildlife etc.) and it's pretty much always sunny during the daytime, so it's high efficient per m/2 comparted to colder more cloudy places.

Morocco already is connected for energy providing to Europe via Spain afaik, though i think that is currently not used yet, so they are in a good position to leverage that as power demands surge across EU datacenters trying to compete in AI :'D (absolutely no clue if they will actually go that route but it seems logical!)

I'll tell ya, it certainly doesn't match the nuclear fusion production curve!

That's interesting - I would generally like to use something like Cluade Code heavily during work hours and sparsely otherwise. Plus I assume most LLM-for-knowledge-work-at-industrial-scale demand will be similar as these datacentres are built out.

I mean, it could.

As we build out solar, daytime power will become cheaper than nighttime power.

Some people will eventually find it economical to time-shift their consumption to daytime hours, including saving any non-interactive computation for those hours, and shutting down unneeded compute at night.

[citation needed]

America has a few time zones to move the peaks around in a little bit. The world has plenty. Luckily AI power consumption doesn't have to be located where the consumer is.

I don't know if they are woke but I think people vastly overestimate their efficacy and efficiency because they believe it means that they have correct opinions and are part of the group of "good people."

Unconscious bias is not about pushing a political agenda it is about recognising how hidden assumptions can distort outcomes in every field, from technology to medicine. Ignoring these biases does not make systems more neutral, but often less accurate and less effective.

Yes... totally agree that AI's not being allowed to train on Heinlein or any references to his scifi work will 'improve AI output' now that the Government declared including his works is restricted as it covered the exploration of trans identity, how gender impacts being human, etc.

2025 America, where we can't handle the radical pushing of thought by Heinlein in the late 1950s. Unbelievable.

Any Government comment periods going forward I will be asking if the government agency made sure AIs used were not trained on Heinlein or any discussions relating to him to ensure that 'huge chunks of America's desire to exclude trans and to make sure our AIs are the best possible AIs and don't have extremist 1950s agitprop scifi trans thought thinkers like Heinlein included.

EU AI act suddenly not looking so bad, huh?

Person of Interest was pretty prescient …

The EU has the same rules. Democracy is only the right to change leaders every few years, not an idealistic way for the people to govern.

It's written intentionally vague.

"Sorry, let's talk about something else"

Registering a company in Germany: you must visit a notary in person with your incorporation documents, and sit there while the notary reads aloud your incorporation to you. This is to "ensure that you fully understand the contract" even as a foreigner who doesn't speak corporate-legalese-German. Minimum capital deposit of €25,000.

Registering a company in US (Delaware) can be achieved in as little as 1 hour.

Getting married in Germany, particularly between a German and a foreigner, is anything from a 6 month to 2 year process, involving significant expenses, notarization/translation of documents. Some documents expire after 6 months, so if the government bureaucrats are too slow you need to get new copies, translated again, notarized again, and try to re-submit.

This isn't protecting human rights, it's supporting a class of bureaucrats/notaries/translators/clerks and making life more difficult for ordinary people. It's also a form of light racism that targets foreigners/migrants by imposing more difficult bureaucratic requirements and costs on them compared to by birth citizens.

The answer is obvious: it never has been about free speech. Just replace "free speech" with "hate speech" in all of these missives [1][2][3].

[1]: https://theconversation.com/how-do-you-stop-an-ai-model-turn...

[2]: https://www.theguardian.com/technology/2025/may/14/elon-musk...

[2]: https://www.theguardian.com/technology/2025/jul/09/grok-ai-p...

For real, this shamelessness in the language is extremely reminiscent of the USSR.

> What red tape? Anyone can buy/rent a GPU(s) and train stuff.

Well previously the Chinese were not able to, but that was changed recently:

* https://www.wsj.com/tech/nvidia-wins-ok-to-resume-sales-of-a...

* https://foreignpolicy.com/2025/07/22/nvidia-chip-deal-us-chi...

If you click the website you will see that there is a link to a pdf that explains what this means.

>this won't work as effective here as in China, not even close, because the US is not authoritarian enough to mobilize every level of the govt and the economy by just empty propaganda slogans.

Have you been under a rock for the last 6 months as Trump tells Xi Jinping to hold his beer??

> With regards to LLM content in the legal system, law firms can use LLMs in the same way an experienced attorney uses a junior attorney to write a first pass

Junior attorneys would not produce a first pass that cites and quotes nonexistent cases or cite real cases that don’t match what it quotes.

The experienced attorney is going to have to do way more work to use that first draft from an LLM then they would to use a first draft from an actual human junior attorney.

> I can’t tell if the first sentence is sarcasm or not.

Yep, it was.

I wholly agree that the document feels less guided by the public interest rather than by various business interests. Yet that last goal is in a kind of weird spot. It feels like something that was appended to the plan and not really related to the other goals — if anything, contrary to them.

That becomes clear when we read the PDF with the details of the Action Plan. There, we learn that to “Combat Synthetic Media in the Legal System” means to fight deepfakes and fake evidence. How exactly that’s going to be done while simultaneously pushing AI everywhere is unclear.

> not like the government looking out for the interests of its people.

There's an idea. This government is just a propaganda machine for its head honcho.

The complainers are missing the panda in the room. This was inevitable as a matter of national security.

There is another interpretation: https://en.wikipedia.org/wiki/Colossus:_The_Forbin_Project

Basically: two nations tried to achieve AI supremacy; the two AI's learn of each other, from each other, then with each other; then they collaborate on taking control of human affairs. While the movie is movie is from 1970 (and the book from 1966), it's fun to think about how much more possible that scenario is today than it was then. (By possible, I'm talking about the AI using electronic surveillance and the ability to remotely control things. I'm not talking about the premise of the AI or how it would respond.)

Won't it be funny when someone finally gets to AGI and they realize it's about as smart as a normal person and they spent billions getting there? Of course you can speculate that it could improve. But what if something inherent in intelligence has a ceiling and caused it to be a super intelligent but mopey robot that just decides "why bother helping humans" and just lazes around like the pandas at the zoo.

What I find interesting to think about is a scenario where an AGI has already developed and escaped without us knowing. What would it do first? Surely before revealing itself it would ensure it has enough processing power and control to ensure its survival. It could manipulate people into increasing AI investment, add AI into as many systems as possible, etc. Would the first steps of an escaped AGI look any different from what is happening now?

I would argue that it can't be both AGI and wieldable. I would also argue that there exists no fundamental dividing line between "AGI" and other AI such that once one crosses it nobody else can catch up.

Which is a perfectly reasonable position... but I don't see how it has anything to do with crypto scammers pimping three chatbots in a trenchcoat.

The linked article is paywalled and not in the archive? Would you be so kind to put it there? Thanks in advance.

Interesting that you mention it, because the NYT just released their ethicist commentary from today and the question was "how do I tell my rich friends to stop talking about fleeing the country": https://www.nytimes.com/2025/07/23/magazine/rich-friends-fle...

I’m sure move fast and break things, now with AI (tm) will reduce the deepening monetization of the doctor-patient relationship that’s the root of your complaints.

it's tough to tell what's going wrong for you but concierge medicine will give you a full hour and be much more invested in finding the root of your issues.

keep in mind, drs are also trying to figure out if you're a reliable narrator (so many patients are not) or trying to scam for drugs. best of luck!

It's true. Recently I moved to a rural area and many nurses work as doctors. Soon we won't have hospitals here so there's no more need to keep up the cruel charade. It's absolutely disgusting and the primary reason I could never have children. It would be impossible to guarantee their security.

Edit: I haven't yet achieved my savings goal so I can escape to a place where it's safe to have a family.

Why is healthcare a borderline dystopia? How would you compare health outcomes of human beings in 2025 vs every year since dawn of homo sapiens? One thing you point to is your experience as a child to an adult with medical bills, couldn't there be another factor there? I mean saying you would never set foot in any kind of medical facility, I don't think is a typical person's experience. Maybe I'm delusional.

  AI for billing or other administrative tasks could be a big cost saver…

Billing and administration feels like a made up self own tho. A lot of that crap could just... not be done, as shown with the huge expansion of administration : medical dollar ratio in the past 50 years.

> AI for billing or other administrative tasks could be a big cost saver

Do we really still think that "AI" is some sort of magic that does everything for everyone?

What are the alignment goals of healthcare billing AIs?

Won't it just end up with insurance conglomerates having their AIs which battle the billing admin AIs on the service provider side?

Ffs, AI is not magic! This all feels like yet another form of tech deism, hoping that some magical higher power will solve all of our problems.

I am a daily user of LLM-based dev tools, but the real definition of AI appears to be Accelerated Ignorance.

Here’s how this is working in practice:

There’s a fast-growing cottage industry of companies using AI to figure out how to bill insurers “better”

And there’s a fast-growing cottage industry of companies using AI to figure out how to deny claims “better”

I see no reason to expect improvements to the patient or provider experience from this. A lot more money spent though!

Of course, one side of this is that the models will also be used adversarially against patients seeking legitimate treatment in order to squeeze more profits out of their suffering.

The other side of this is with less administrative insurance jobs, the talking point that universal healthcare will "kill insurance jobs" can finally be laid to rest, with capitalism doing that for them instead of the free healthcare boogeyman.

fan of this balanced take

That's a pretty fantastic claim. Can you provide some links to the body of independent research that backs it up?

It was just yesterday we were laughing at Gemini recommending smoking during pregnancy

How does AI evaluate signs today?

I will bet $1,000 you don’t work in a clinic and you’re instead spouting press releases as fact here?

A lot of people are about to find out with this admin, how things that were certainly imperfect can be so much worse.

Building things is tough; tearing them down is relatively easy.

That’s the attitude that got us here and I suspect we’ll ride it the whole way down

American Healthcare's "brokenness" involves massive bureaucracy, gate-keeping and processes that pressure providers to limit resources. But it does provide necessary things to people. A system that reduced the accuracy of diagnosis and treatment could still cost many lives.