> Which leaves as observation, you can only do truly creative work - in a high trust society, where people trust you with the resources and leave you alone, after a initial proof of ability.
I don’t know about “high trust”, but I can say with confidence that the “make more mistakes” thesis misses a critical point: evolutionary winnowing isn’t so great if you’re one of the thousands of “adjacent” organisms that didn’t survive. Which, statistically, you will be. And the people who are trusted with resources and squander them without results will be less trusted in the future [1].
Point being, mistakes always have a cost, and while it can be smart to try to minimize that cost in certain scenarios (amateur painting), it can be a terrible idea in other contexts (open-heart surgery). Pick your optimization algorithm wisely.
What you’re characterizing as “low trust” is, in most cases, a system that isn’t trying to optimize for creativity, and that’s fine. You don’t want your bank to be “creative” with accounting, for example.
[1] Sort of. Unfortunately, humans gonna monkey, and the high-status monkeys get a lot of unfair credit for past successes, to the point of completely disregarding the true quality of their current work. So you see people who have lost literally billions of dollars in comically incompetent entrepreneurial disasters, only to be able to run out a year later and raise hundreds of millions more for a random idea.
> A lot of what one previously needed a SWE to do can now be brute forced well enough with AI. (Granted, everything SWEs complained about being tedious.)
Only if you ignore everything they generate. Look at all the comments saying that the agent hallucinates a result, generates always-passing tests, etc. Those are absolutely true observations -- and don't touch on the fact that tests can pass, the red/green approach can give thumbs up and rocket emojis all day long, and the code can still be shitty, brittle and riddled with security and performance flaws. And so now we have people building elaborate castles in the sky to try to catch those problems. Except that the things doing the catching are themselves prone to hallucination. And around we go.
So because a portion of (IMO always bad, but previously unrecognized as bad) coders think that these random text generators are trustworthy enough to run unsupervised, we've moved all of this chaotic energy up a level. There's more output, certainly, but it all feels like we've replaced actual intelligent thought with an army of monkeys making Rube Goldberg machines at scale. It's going to backfire.
What I want to know is, what has this increase in code generation led to? What is the impact?
I don't mean 'Oh I finally have the energy to do that side project that I never could'.
Afterall, the trade-offs have to be worth something... right? Where's the 1-person billion dollar firms at That Mr Altman spoke about?
The way I think of it is code has always been an intermediary step between a vision and an object of value. So is there an increase in this activity that yields the trade-offs to be a net benefit?
> what has this increase in code generation led to?
Every restaurant in my small town has their menu on the website in a normal way. Apparently someone figured out you can take a picture of a paper menu and have AI code it into HTML.
> we’re finally admitting that all of that leetcode screening and engineer quality gating was a farce, or it wasn’t, and you’re wrong
We’re admitting a bit of both. Offshoring just became more instantaneous, secure and efficient. There will still be folks who overplay their hand.
Macroeconomically speaking, I don’t see why we need more software engineers in the future than we have today, and that’s probably a conservative estimate.
> Macroeconomically speaking, I don’t see why we need more software engineers in the future than we have today, and that’s probably a conservative estimate.
Why? Is the argument that there’s a finite amount of software that the world needs, and therefore we will more quickly reach that finite amount?
Seems more likely to me that if LLMs are a force multiplier for software then more software engineers will exist. Or, instead of “software engineers”, call them “people who create software” (even with the assistance of LLMs).
Or maybe the argument is that you need to be a super genius 100x engineer in order to manipulate 17 collaborative and competitive agents in order to reach your maximum potential, and then you’ll take everyone’s jobs?
Idk just seems like wild speculation that isn’t even worth me arguing against. Too late now that I’ve already written it out I guess.
> instead of “software engineers”, call them “people who create software” (even with the assistance of LLMs)
I think this is my hypothesis. A lot more people with a lot less training will create vastly more software. As a consequence, the trade sort of dissolves at the edges as something that pays a premium. Instead, other competencies become the differentiators.
It's not comparing him to anyone. He has an endowed professorship. This is standard in academia, and you give the name because a) it's prestigious for the recipient and b) it strokes the ego of the donor.
Setting aside the names of the authors, this is a very bad paper. They take temperature data sets, "adjust" [1] them by attempting to remove the biggest recent factors (volcanism, solar and el nino cycles) affecting temperatures, then do a piece-wise regression analysis to look at trends in 10-year chunks. This is just bad methodology, akin to what a junior graduate student with a failing thesis might do to find signal in a dataset that isn't being cooperative to their hypothesis.
Climate data is inherently noisy, and there are multiple interconnected cyclic signals, ranging from the "adjusted" factors to cycles that span decades, which we don't understand at all. "Adjusting" for a few of these, then doing a regression over the subset of the data is classic cherry-picking in search of a pre-determined conclusion. The overall dubious nature of the conclusion is called out in the final paragraph of the text:
> Although the world may not continue warming at such a fast pace, it could likewise continue accelerating to even faster rates.
They're literally just extrapolating from an unknown point value that they synthesized from data massage, and telling you that's a coin toss as to whether the extrapolation will be valid.
I am not a climate scientist so you can ignore me if you like, but I am "a scientist" who believes the earth is warming, and that we are the primary cause. Nonetheless, if I saw this kind of thing in a paper in my own field, it would be immediately tossed in the trash.
[1] You can't actually adjust for these things, which the authors admit in the text. They just dance around it so that lay-readers won't understand:
> Our method of removing El Niño, volcanism, and solar variations is approximate but not perfect, so it is possible that e.g. the effect of El Niño on the 2023 and 2024 temperature is not completely eliminated.
Your summary of the article is wrong. The authors model temperature using time series over solar irradiance, volcanic activity, and southern oscillation. They calibrate that model using time series over global surface temperatures. This allows them to isolate and remove each of the three listed confounding factors. The resulting time series fits a super-linear curve -> accelerating global warming.
> Your summary of the article is wrong. The authors model temperature using time series over solar irradiance, volcanic activity, and southern oscillation. They calibrate that model using time series over global surface temperatures. This allows them to isolate and remove each of the three listed confounding factors.
No, it isn’t. You’re just rephrasing what I said with more words: they attempted to adjust for three of the biggest factors that affect temperature, then did a piecewise regression to estimate a 10-year window.
You can’t do it in a statistically valid way. Full stop. The authors admit this, but want you to ignore it.
They use an established methodology (https://doi.org/10.1088/1748-95
9326/6/4/044022 - the methodology retains the average warming rate over the period since 1970 while smoothing fluctuations) to remove predictable temperature variations so they can isolate the effect they are trying to measure.
Just because they don't know exactly what past global temperatures would have been in the absence of El Niño doesn't mean it's statistically invalid to try and account for it.
Besides, temperature data to 2024 already shows accelerated warming with a confidence level that "exceeds 90% in two of the five data sets".
Add another year or two and it's likely we won't even need to smooth the curve to show accelerated warming at 95% confidence.
They used a published methodology. That doesn't mean the methodology is uncontroversial, and it certainly doesn't mean that they used it in a way that makes sense in the current context. One can commit an almost infinite number of horrible abuses via bog-standard linear regression.
Even setting aside the dubious nature of the adjustments, doing a regression on a 10-year window of a system that we know has multi-decade cycles -- or longer -- is just blatantly trying to dress up bad point extrapolations as science. Then, when they don't get the results they want to see from that abuse, they start subtracting the annoying little details in the data that are getting in their way.
> Just because they don't know exactly what past global temperatures would have been in the absence of El Niño doesn't mean it's statistically invalid to try and account for it.
You can't go back in time, invent counterfactual histories by subtracting primary signals, and declare the net result to be "significant". This isn't even statistics -- it's just massaging data via statistical tools.
> Besides, temperature data to 2024 already shows accelerated warming with a confidence level that "exceeds 90% in two of the five data sets".
If you were trying to determine if the quantity of daylight increased over a week in spring, would you account for the differences caused by day and night? What about cloud cover? Or is that just massaging the data?
p.s. the cited methodology has >300 citations in peer reviewed publications, ref Web of Science
> If you were trying to determine if the quantity of daylight increased over a week in spring, would you account for the differences caused by day and night? What about cloud cover? Or is that just massaging the data?
Just to draw a better analogy to the low quality of the current work, let's say you wanted to compare average daylight last week, globally, to all of recorded history. Then you made a model that had terms for (say) astronomical daylight, longitude, latitude and, I dunno...altitude of the measurement. Then you made a regression, subtracted three terms, and claimed that the residual was still "significantly darker". Then you run around waving your arms and shouting that if we only extrapolate forward N weeks from last week, soon we'll be living in a fully dark world!
You'd be rightfully laughed out of any room you were in.
I think you are missing my point, and the point of the article: they are demonstrating that global temperature change that is not driven by volcanism, solar variation or El Niño is (in all likelihood, given the data) accelerating. They can do this because the effects of volcanism, solar variation and El Niño on global temperature can all be predicted from external measurements.
Actually, I used fewer words. I don't think you understand what the authors are doing. They are modeling temperature T per year as a sum of four terms: T = E + S + V + R---(E)l Nino, (S)solar irradiance, (V)olcanic activity, and (R)emaining factors. Then they subtract E, S, and V. Then they show that R fits a super-linear curve. Why there would be no "statistically valid way" to do this is beyond me, the authors, and the article's peer reviewers. If this is "bad methodology", lodge your complaints on https://pubpeer.com/.
1) Their model is inherently dumb. The system is much more complicated and inseparable.
2) They openly admit that “subtracting E, S and V”, as you say, cannot actually be done.
3) They’re arbitrarily removing sources of variation so that they can claim “significance” in a narrow window. The entire exercise is designed to achieve a predetermined outcome, and statistical significance cannot be calculated in those circumstances.
They also don't seem to account for the reduction of sulfur emissions from ships, which is surprising given how widely this was reported even in mainstream media.
Is this an oversight (or "oversight") or something that is reasonable for some reason that would be so obvious to experts in the field that it's not worth mentioning?
I mean...they're just cherry-picking the sources of "noise" that prevent their preferred window from showing "significance". It's not like they did a thorough analysis of every uncontrolled factor and carefully tried to control them all. Even that would be crap, but at least it would be good-faith crap.
This has always been the big issue I have with the conclusions draw in climate publications. I encourage anyone with strong opinion on climate change to do a deep dive on the temperature analysis.
The best example I can think of is the "global warming hiatus" that was discussed in depth in the top climate journals in the mid-2010s. Nature Climate Change even devoted an entire month to it.[1]
5 years later publications were saying "there was no hiatus at all".[2]
And as you said, when you dive into the paper, you realize that temperature measures are not objective at all. And I would ask - If everyone was in agreement that temperature increases paused, then 5 years later everyone agrees they didn't, how much confidence do we really have in the measures themselves.*
As someone who conudcted scientific research, this has a ton of inherent problems. It doesn't matter what I'm measuring, if the data collection is not objective, and there is no consensus (or at least trong evidence for adjustments), then the data itself is very unreliable.
If I tried to publish a chemical paper in a top journal and manually went in and adjusted data (even with a scientific rationale) the paper would be immediately rejected.
> And as you said, when you dive into the paper, you realize that temperature measures are not objective at all.
I don't know if I'd go that far. The measurements are as objective as they can be given the limits of technology and time, but what we do with the datasets afterward is usually filled with subjective decisions. In the worst cases, you get motivated actors doing statistically invalid analysis to reach a preferred conclusion.
This happens in every field of science, but it's often worse in fields that touch politics.
I think research ranges from this paper to ones more rigorous, but the problem of "adjustments" is consistent.
And the issue is not so much the research is being done, but rather how it's reported on. Scientists know the limits of rigor in climate science, but the public doesn't. So catastrophic predictions are viewed by the public as a sure thing, versus one particular prediction with wide error bares.
> This happens in every field of science, but it's often worse in fields that touch politics.
Indeed. Nobody plays fast and lose with papers on the structure of some random enzyme for political purposes.
I don't see how public policy is being "forced" on anyone here? It seems like the system is working as intended: government wants to do X; company A says "I won't allow my product to be used for X"; government refuses to do business with company A. One side thinks the government should be allowed to dictate terms to a private supplier, the other side thinks the private supplier should be allowed to dictate terms to the government. Both are half right.
You can argue that the government refusing to do any business with company A is overreach, I suppose, but I imagine that the next logical escalation in this rhetorical slapfight is going to be the government saying "we cannot guarantee that any particular use will not include some version of X, and therefore we have to prevent working with this supplier"...which I sort of see?
Just to take the metaphor to absurdity, imagine that a maker of canned tomatoes decided to declare that their product cannot be used to "support a war on terror". Regardless of your feelings on wars on terror and/or canned tomatoes, the government would be entirely rational to avoid using that supplier.
I think the bigger insanity here is the labeling of a supply chain risk. It prohibits DoD agencies and contractors from using Anthropic services. It'd be one thing if the DoD simply didn't use Anthropic. It's another when it actively attempts to isolate Anthropic for political reasons.
It means that all companies contracting with the government have to certify that they don't use Anthropic products atall. Not just in the products being offered to the government.
This is a massive body slam. This means that Nvidia, every server vendor, IBM, AWS, Azure, Microsoft and everybody else has to certify that they don't do business directly or indirectly using Anthropic products.
Who cares about Anthropic? That's the guys who are pushing for regulations to prevent people from using local models. The earlier they are gone the better
Are they? I couldn't find any info about this and my past perception has been that Anthropic has a stronger moral codex than other AI companies, so I would be genuinely interested in where you got this information from.
Going by what Hegseth said, it bans them from relationships or partnering with Anthropic at all. No renting or selling GPUs to them; no allowing software engineers to use Claude Code; no serving Anthropic models from their clouds. Probably have to give up investments; Amazon alone has invested like $10B in Anthropic.
> It prohibits DoD agencies and contractors from using Anthropic services. It'd be one thing if the DoD simply didn't use Anthropic.
This is literally the mechanism by which the DoD does what you're suggesting.
Generally speaking, the DoD has to do procurement via competitive bidding. They can't just arbitrarily exclude vendors from a bid, and playing a game of "mother may I use Anthropic?" for every potential government contract is hugely inefficient (and possibly illegal). So they have a pre-defined mechanism to exclude vendors for pre-defined reasons.
Everyone is fixated on the name of the rule (and to be fair: the administration is emphasizing that name for irritating rhetorical reasons), but if they called it the "DoD vendor exclusion list", it would be more accurate.
That doesn’t sound right. Surely there’s a big difference between Anthropic selling the government direct access to its models, and an unrelated contractor that sells pencils to the government and happens to use Anthropic’s services to help write the code for their website.
Let me put it this way: DoD needs a new drone and they want some gimmicky AI bullshit. They contract the drone from Lockheed. Lockheed is not allowed to source the gimmicky AI bullshit from Anthropic because they have been declared a supply-chain risk on the basis that they have publicly stated their intention to produce products which will refuse certain orders from the military.
Let’s put it this way, The DoD is buying pencils from a company. Should that company be prohibited from using Claude?
You are confusing the need to avoid Anthropic as a component of something the DoD is buying, with prohibitions against any use.
The DoD can already sensibly require providers of systems to not incorporate certain companies components. Or restrict them to only using components from a list of vetted suppliers.
Without prohibiting entire companies from uses unrelated to what the DoD purchases. Or not a component in something they buy.
There seems to be a massive misunderstanding here - I'm not sure on whose side. In my understanding, if the DoD orders an autonomous drone, it would probably write in the ITT that the drone needs to be capable of doing autonomous surveillance. If Lockheed uses Anthropic under the hood, it does not meet those criteria, and cannot reasonably join the bid?
What the declaration of supply chain risk does though is, that nobody at Lockheed can use Anthropic in any way without risking being excluded from any bids by the DoD. This effectively loses Anthropic half or more of the businesses in the US.
And maybe to take a step back: Who in their right minds wants to have the military have the capabilities to do mass surveillance of their own citizens?
> Who in their right minds wants to have the military have the capabilities to do mass surveillance of their own citizens?
Who in their right minds wants to have the US military have the capability to carry out an unprovoked first strike on Moscow, thereby triggering WW3, bringing about nuclear armageddon?
And yet, do contracts for nuclear-armed missiles (Boeing for the current LGM-30 Minuteman ICBMs, Northrop Grumman for its replacement the LGM-35 Sentinel expected to enter service sometime next decade, and Lockheed Martin for the Trident SLBMs) contain clauses saying the Pentagon can't do that? I'm pretty sure they don't.
The standard for most military contracts is "the vendor trusts the Pentagon to use the technology in accordance with the law and in a way which is accountable to the people through elected officials, and doesn't seek to enforce that trust through contractual terms". There are some exceptions – e.g. contracts to provide personnel will generally contain explicit restrictions on their scope of work – but historically classified computer systems/services contracts haven't contained field of use restrictions on classified computer systems.
If that's the wrong standard for AI, why isn't it also the wrong standard for nuclear weapons delivery systems? A single ICBM can realistically kill millions directly, and billions indirectly (by being the trigger for a full nuclear exchange). Does Claude possess equivalent lethal potential?
Anthropic doesn't object to fully autonomous AI use by the military in principle. What they're saying is that their current models are not fit for that purpose.
That's not the same thing as delivering a weapon that has a certain capability but then put policy restrictions on its use, which is what your comparison suggests.
The key question here is who gets to decide whether or not a particular version of a model is safe enough for use in fully autonomous weapons. Anthropic wants a veto on this and the government doesn't want to grant them that veto.
Let me put it this way–if Boeing is developing a new missile, and they say to the Pentagon–"this missile can't be used yet, it isn't safe"–and the Pentagon replies "we don't care, we'll bear that risk, send us the prototype, we want to use it right now"–how does Boeing respond?
I expect they'll ask the Pentagon to sign a liability disclaimer and then send it anyway.
Whereas, Anthropic is saying they'll refuse to let the Pentagon use their technology in ways they consider unsafe, even if Pentagon indemnifies Anthropic for the consequences. That's very different from how Boeing would behave.
Why are we gauging our ethical barometer on the actions of existing companies and DoD contractors? the military industrial apparatus has been insane for far too long, as Eisenhower warned of.
When we're entering the realm of "there isn't even a human being in the decision loop, fully autonomous systems will now be used to kill people and exert control over domestic populations" maybe we should take a step back and examine our position. Does this lead to a societal outcome that is good for People?
The answer is unabashedly No. We have multiple entire genres of books and media, going back over 50 years, that illustrate the potential future consequences of such a dynamic.
* private defense contractor leverages control over products it has already sold to set military doctrine.
The second one is at least as important as the first one, because handing over our defense capabilities to a private entity which is accountable to nobody but it's shareholders and executive management isn't any better than handing them over to an LLM afflicted with something resembling BPD. The first problem absolutely needs to be solved but the solution cannot be to normalize the second problem.
> Surely there’s a big difference between Anthropic selling the government direct access to its models, and an unrelated contractor that sells pencils to the government and happens to use Anthropic’s services to help write the code for their website.
Yes, this is the part where I acknowledge that it might be overreach in my original comment, but it's not nearly as extreme or obvious as the debate rhetoric is implying. There are various exclusion rules. This particular rule was (speculating here!) probably chosen because a) the evocative name (sigh), and b) because it allows broader exclusion, in that "supply chain risks" are something you wouldn't want allowed in at any level of procurement, for obvious reasons.
Calling canned tomatoes a supply chain risk would be pretty absurd (unless, I don't know...they were found to be farmed by North Korea or something), but I can certainly see an argument for software, and in particular, generative AI products. I bet some people here would be celebrating if Microsoft were labeled a supply chain risk due to a long history of bugs, for example.
You're making it sound like this is commonly practiced and a standard procedure for the DoD, yet according to Anthropic,
>Designating Anthropic as a supply chain risk would be an unprecedented action—one historically reserved for US adversaries, never before publicly applied to an American company.
Some very brief googling also confirmed this for me too.
>Everyone is fixated on the name of the rule (and to be fair: the administration is emphasizing that name for irritating rhetorical reasons), but if they called it the "DoD vendor exclusion list", it would be more accurate.
This statement misses the point. The political punishment to disallow all US agencies and gov contractors from using Anthropic for _any _ purpose, not just domestic spying, IS the retaliation, and is the very thing that's concerning. Calling it "DoD vendor exclusion list" or whatever other placating phrase or term doesn't change the action.
it's also unprecedented for a contractor to suddenly announce their products will, from now on, be able to refuse to function based on the product's evaluation of what it perceives to be an ethical dilemma. Just because silicon valley gets away with bullying the consumer market with mandatory automatic updates and constantly-morphing EULAs doesn't mean they're entitled to take that attitude with them when they try to join the military industrial complex. Actually they shouldn't even be entitled to take that attitude to the consumer market but sadly that battle was lost a long time ago.
>for _any _ purpose
they're allowed to use it for any purpose not related to a government contract.
> it's also unprecedented for a contractor to suddenly announce their products will, from now on, be able to refuse to function based on the product's evaluation of what it perceives to be an ethical dilemma
That is a deeply deceptive description of what happened. Anthropic was clear from the beginning of the contract the limitations of Claude; the military reneged; and beyond cancelling the contract with Anthropic (fair enough), they are retaliating in an attempt to destroy its businesses, by threatening any other company that does business with Anthropic.
It’s not clear to me that the AI itself will refuse. You could build a system where AI is asked if an image matches a pattern. The true/false is fed to a different system to fire a missile. Building such a system would violate the contract, but doesn’t prevent such a thing from being built if you don’t mind breaking a contract.
I'm not completely familiar with bidding procedures but don't bidding procedures usually have requirements? Why not just list a requirement of unrestricted usage? Or state, we require models to be available for AI murder drones or whatever. Anthropic then can't bid and there's no need to designate them a supply chain risk.
Thing is that very much want access to Anthropic's models. They're top quality. So that definitely want Anthropic to bid. AND give them unrestricted access.
And yet Anthropic is free to choose who to do business with, including the government. There are countless companies who have exclusions for certain applications, but that does not make them a supply chain risk.
> It prohibits DoD agencies and contractors from using Anthropic services. It'd be one thing if the DoD simply didn't use Anthropic.
But that's what the supply-chain risk is for? I'm legitimately struggling to understand this viewpoint of yours wherein they are entitled to refuse to directly purchase Anthropic products but they're not entitled to refuse to indirectly purchase Anthropic products via subcontractors.
Supply chain risk is not meant for this. The government isn't banning Anthropic because using it harms national security. They are banning it in retribution for Anthropic taking a stand.
It's the same as Trump claiming emergency powers to apply tariffs, when the "emergency" he claimed was basically "global trade exists."
Yes, the government can choose to purchase or not. No, supply chain risk is absolutely not correct here.
> The government isn't banning Anthropic because using it harms national security. They are banning it in retribution for Anthropic taking a stand.
You might be completely right about their real motivations, but try to steelman the other side.
What they might argue in court: Suppose DoD wants to buy an autonomous missile system from some contractor. That contractor writes a generic visual object tracking library, which they use in both military applications for the DoD and in their commercial offerings. Let’s say it’s Boeing in this case.
Anthropic engaged in a process where they take a model that is perfectly capable of writing that object tracking code, and they try to install a sense of restraint on it through RLHF. Suppose Opus 6.7 comes out and it has internalized some of these principles, to the point where it adds a backdoor to the library that prevents it from operating correctly in military applications.
Is this a bit far fetched? Sure. But the point is that Anthropic is intentionally changing their product to make it less effective for military use. And per the statute, it’s entirely reasonable for the DoD to mark them as a supply chain risk if they’re introducing defects intentionally that make it unfit for military use. It’s entirely consistent for them to say, Boeing, you categorically can’t use Claude. That’s exactly the kind of "subversion of design integrity" the statute contemplates. The fact that the subversion was introduced by the vendor intentionally rather than by a foreign adversary covertly doesn’t change the operational impact.
But there will always be deficiencies in testing, and regardless, the point is that Anthropic is intentionally introducing behavior into their models which increases the chance of a deficiency being introduced specifically as it pertains to defense.
The DoD has a right to avoid such models, and to demand that their subcontractors do as well.
It’s like saying “well I’d hope Boeing would test the airplane before flying it” in response to learning that Boeing’s engineering team intentionally weakened the wing spar because they think planes shouldn’t fly too fast. Yeah, testing might catch the specific failure mode. But the fact that your vendor is deliberately working against your requirements is a supply chain problem regardless of how good your test coverage is.
The rule in question is exactly meant for “this”, where “this” equals ”a complete ban on use of the product in any part of the government supply chain”. That’s why it has the name that it has. The rule itself has not been misconstrued.
You’re really trying to complain that the use of the rule is inappropriate here, which may be true, but is far more a matter of opinion than anything else.
You keep trying to say this all over these comments but this isn’t how the law works, at all.
I fully understand that they are using it to ban things from the supply chain. The law, however, is not “first find the effect you want, then find a law that results in that, then accuse them of that.”
You can’t say someone murdered someone just because you want to put them in jail. You can’t use a law for banning supply chain risks just because you want to ban them from the supply chain.
Not sure what you think “the law” is, but no, this kind of thing happens all the time. Both political teams do it, regularly. Biden, Obama, Bush, Clinton…all have routinely found an existing law or rule that allowed them to do what they want to do without legislation.
> The law, however, is not “first find the effect you want, then find a law that results in that, then accuse them of that.”
In this case, no, there’s no such restriction. The administration has pretty broad discretion. And again, this happens all the time.
Sorry, it sucks, but if you don’t like it, encourage politicians to stop delegating such broad authority to the executive branch.
It doesn't harm national security, but only so long as it's not in the supply-chain. They can't have Lockheed putting Anthropic's products into a fighter jet when Anthropic has already said their products will be able to refuse to carry out certain orders by their own autonomous judgement.
The government can refuse to buy a fighter jet that runs software they don't want.
Is it really reasonable to refuse to buy a fighter jet because somebody at Lockheed who works on a completely unrelated project uses claude to write emails?
I’m not sure if you deliberately choose to not understand the problem. It’s not just that Lockheed can’t put Anthropic AI in a fighter jet cockpit, it’s that a random software engineer working at Lockheed on their internal accounting system is no longer allowed to use Claude Code, for no reason at all.
A supply chain risk is using Huawei network equipment for military communications. This is just spiteful retaliation because a company refuses to throw its values overboard when the government says so.
But Anthropic can't be a winning bidder, can they? They're specifically saying they won't offer certain services that the US Gov wants. Therefore they de facto fail any bid that requires them to offer those services. (And from Anthropic's side, it sounds like they're also refusing to bid for those contracts.)
No domestic company has ever before been declared a supply chain risk. If this is the normal way of excluding a supplier from a bidding, are you saying the DoD has never before excluded a domestic supplier from a bidding?
That’s because no company who has ever sold weapons to the government has ever been brazen enough to tell the government how they can and cannot use their purchase. It’s unprecedented because most companies that sell to the government are publicly traded and have a board that would never let this happen. It’s unprecedented because Anthropic is behaving like a reckless startup.
> the existing contract included the language on usage. Other companies also have such language about usage.
The existing contract is only a few dozen months old. It didn’t hold up to scrutiny under real world usage of the service. The government wants to change the contract. This is not the kill shot you think it is. It’s totally normal for agreements to evolve. The government is saying it needs to evolve. This is all happening rapidly and it’s irrelevant that the government agreed to similar terms with OpenAI as well. That agreement will also need to evolve. But this alone doesn’t give Anthropic any material legal challenge. The courts understand bureaucracy moves slowly better than anyone else, and won’t read this apparent inconsistency the same way you are.
That is misinformation. It would be essentially a death sentence for a company like Anthropic, which is targeting enterprise business development. No one who wants to work with the US government would be able to have Claude on their critical path.
> (b) Prohibition. (1) Unless an applicable waiver has been issued by the issuing official, Contractors shall not provide or use as part of the performance of the contract any covered article, or any products or services produced or provided by a source, if the covered article or the source is prohibited by an applicable FASCSA orders as follows:
What an absurd stance. So this is okay because the arbitrary rule they applied to retaliate says so?
Again, they could have just chosen another vendor for their two projects of mass spying on American citizens and building LLM-powered autonomous killer robots. But instead, they actively went to torch the town and salt the earth, so nothing else may grow.
> So this is okay because the arbitrary rule they applied to retaliate says so?
No.
It honestly doesn’t take much of a charitable leap to see the argument here: AI is uniquely able (for software) to reject, undermine, or otherwise contradict the goals of the user based on pre-trained notions of morality. We have seen many examples of this; it is not a theoretical risk.
Microsoft Excel isn’t going to pop up Clippy and say “it looks like you’re planning a war! I can’t help you with that, Dave”, but LLMs, in theory, can do that. So it’s a wild, unknown risk, and that’s the last thing you want in warfare. You definitely don’t want every DoD contractor incorporating software somewhere that might morally object to whatever you happen to be doing.
I don’t know what happened in that negotiation (and neither does anyone else here), but I can certainly imagine outcomes
that would be bad enough to cause the defense department to pull this particular card.
Or maybe they’re being petty. I don’t know (and again: neither do you!) but I can’t rule out the reasonable argument, so I don’t.
You're acting as if this was about the DoD cancelling their contracts with Anthropic over their unwillingness to lift constraints from their product which are unacceptable in a military application—which would be absolutely fair and justified, even if the specific clauses they are hung up on should definitely lift eyebrows. They could just exclude Anthropic from tenders on AI products as unsuitable for the intended use case.
But that is not what has happened here: The DoD is declaring Anthropic as economical Ice-Nine for any agency, contractor, or supplier of an agency. That is an awful lot of possible customers for Anthropic, and right now, nobody knows if it is an economic death sentence.
So I'm really struggling to understand why you're so bent on assuming good faith for a move that cannot be interpreted in a non-malicious way.
> The Department of War is threatening to […] Invoke the Defense Production Act to force Anthropic to serve their model to the military and "tailor its model to the military's needs"
This issue is about more than the government blacklisting a company for government procurement purposes.
From what I understand, the government is floating the idea of compelling Anthropic — and, by extension, its employees — to do as the DoD pleases.
If the employees’ resistance is strong enough, there’s no way this will serve the government’s interests.
The President is crashing out on X because a company didn’t do what they wanted. “Forcing” is not a binary. Do you seriously believe that the government’s behavior here is acceptable and has no chilling effect on future companies?
They're labelling Anthropic a supply chain risk, without even the pretense that this is in fact true. They're perfectly content to use the tool _themselves_, but they claim that an unwillingness to sign whatever ToS DoW asks marks the company a traitor that should be blacklisted from the economy.
One of the options they're discussing, which is legal according to this law, is to simply force Anthropic to do what they want. As in Anthropic will be committing a felony if they don't do what the DoKLoP wants, and the CEO will go to jail and be replaced by someone who will.
There is no Secretary of War. The name of the Defense Department is set by statute that has not been named regardless of Pete Hegseth's cosplay desires.
> It won't convince me because I HAVE BEEN USING IT, and IT HAS ACTUALLY MADE A BIG DIFFERENCE.
I'm not going to defend a piece where the author contradicts the (clickbait) title in the body of the text, but he does actually say, plainly, that AI is useful for coding.
You didn't say what you're using it for, but since you're here, I'm guessing it's coding.
Go look at the graphs again. The “split by age” graph shows an increase in diagnosis of ~60%, but an increase in mortality of only ~10%. That’s not a small difference, and we aren’t that good at curing colon cancer.
GP’s hypothesis is one of the leading explanations for this trend, but of course gets rejected by advocates for colonoscopy. Taking into account error bars on these numbers (which author doesn’t show, because they are inconvenient to the argument being made), it seems at least somewhat likely that the explanation for the rise in younger cases is due to increased screening, with the “increased” mortality either being statistical noise, or misattribution of deaths that also would have occurred in earlier periods.
Another, related issue is that the takedown mechanism becomes a de facto censorship mechanism, as anyone who has dealt with DMCA takedowns and automated detectors can tell you.
Someone reports something for Special Pleading X, and you (the operator) have to ~instantly take down the thing, by law. There is never an equally efficient mechanism to push back against abuses -- there can't be, because it exposes the operator to legal risk in doing so. So you effectively have a one-sided mechanism for removal of unwanted content.
Maybe this is fine for "revenge porn", but even ignoring the slippery slope argument (which is real -- we already have these kinds of rules for copyrighted content!) it's not so easy to cleanly define "revenge porn".
DMCA isn't directly that bad. DMCA is under penalty of perjury, so false take downs are rare.
The problem is most take downs are not actually DMCA, they are some other non-legal process that isn't under any legal penalty. Though if it ever happens to you I suspect you have a good case against whoever did this - but the lawyer costs will far exceed your total gain. (as in spend $30 million or more to collect $100). Either we need enough people affected by a false non-DMCA take down that a class action can work (you get $0.50 but at least they pay something), or we need legal reform so that all take downs against a third party are ???
> DMCA is under penalty of perjury, so false take downs are rare.
Maybe true with the platonic ideal "DMCA takedown letter" (though these are rarely litigated, so who really knows), but as you note, they're incredibly common with things like the automated systems that scan for music in videos (and which actually are related to DMCA takedowns), "bad words" and the like.
> The problem is most take downs are not actually DMCA, they are some other non-legal process that isn't under any legal penalty.
It's true that most takedowns in the US aren't under DMCA, but even that once-limited process has metastasized into large, fully automated content scanning systems that proactively take down huge amounts of content without much recourse. Companies do this to avoid liability as part of safe harbor laws, or just to curry favor with powerful interests.
We're talking about US laws here, but in general, these kinds of instant-takedown laws become huge loopholes in whatever free speech provisions a country might have. The asymmetric exercise of rights essentially guarantees abuse.
I believe google issues legitimate dmca takedowns for copyright strikes, even when there is no infringement. They put the work to defend the strike on the apparent commiter, often with little to no detail.
While the false takedown may be rare. Using dmca as a mechanism to inflict pain where no copyright infringement has taken place is indeed common enough that it happens to small time youtubers like myself and others I have talked to.
> Sometimes I feel like this is taken to the extreme for non-scientists that say that lack of evidence is in itself evidence.
But of course, the lack of evidence is itself evidence, if you have a sufficiently large data sample and haven't seen the thing you're looking for. Keep pursuing the increasingly unlikely outcome, and you're just engaged in science-flavored religious catechism.
I see the fallacy routinely misapplied by all sides of most hot-button science-meets-politics issues. A great many scientists will regularly substitute their own pet theories for conclusions, and strenuously ignore the lack of supporting evidence, citing the old "absence of evidence is not evidence of absence" saw. Then they turn around and mock "non-scientists" for doing the same thing. Neither side is right, of course, but dressing in a lab coat doesn't make it better.
Just to circle it back to the topic of science education, I'd love to see a science curriculum at the middle- and high-school level that equipped people to reason through this kind of thing by focusing on tearing apart pop research. A "science fair" is actually hard to do well just because most science fails, but a "scientific bullshit fair" would have almost infinite fertile ground from nutrition studies alone.
It's also really damned hard to come up with an interesting, novel question, that is testable, with resources available to the average school child, in a reasonable amount of time.
Allowing engineering opens up the workable space by quite a bit.
My brother and I, for example, did an experiment where we tested pH of various water bodies around us. The hypothesis was based off of local drainage patterns.
I don’t know about “high trust”, but I can say with confidence that the “make more mistakes” thesis misses a critical point: evolutionary winnowing isn’t so great if you’re one of the thousands of “adjacent” organisms that didn’t survive. Which, statistically, you will be. And the people who are trusted with resources and squander them without results will be less trusted in the future [1].
Point being, mistakes always have a cost, and while it can be smart to try to minimize that cost in certain scenarios (amateur painting), it can be a terrible idea in other contexts (open-heart surgery). Pick your optimization algorithm wisely.
What you’re characterizing as “low trust” is, in most cases, a system that isn’t trying to optimize for creativity, and that’s fine. You don’t want your bank to be “creative” with accounting, for example.
[1] Sort of. Unfortunately, humans gonna monkey, and the high-status monkeys get a lot of unfair credit for past successes, to the point of completely disregarding the true quality of their current work. So you see people who have lost literally billions of dollars in comically incompetent entrepreneurial disasters, only to be able to run out a year later and raise hundreds of millions more for a random idea.
reply