Honestly I think you nailed the core absurdity here. We're racing China on AI because we don't want to live in a country where the state can destroy any private company that won't do what it's told. And then the Pentagon dusts off a Korean War ammo factory law to coerce a chatbot company in San Francisco. You can't make this up.
The surveillance math is the part that should scare people more than it does. $30B today, pocket change by 2030. At that point the only thing between us and full panopticon is the political expectation that we don't do that here. Which is a norm, not a law. And norms only exist if someone is willing to take a hit to defend them. That's what Anthropic is doing right now, whether or not it works.
nobody in the "yay Anthropic" camp wants to sit with corporate red lines are moot within 18 months once open source catches up. Some fine-tuned Llama descendant will happily notetake a camera feed for whoever asks. The fix has to be political. There is no technical solution to a political problem.
Filing this under essays I'll be linking back to in 3 years when everyone pretends these questions are new.
If a supplier for some commodity within the supply chain of the US military makes it well known that they will discontinue service to any intermediary who violates usage terms that the military itself finds unacceptable, how does that not put the supply chain at risk?
You realize department of war is not a thing right? It takes an act of Congress to rename which hasn't happened so you just sound like someone calling the Gulf of Mexico the Gulf of America
Just fyi your picture for Stanislav Petrov is incorrect. The picture you showed is of Vasily Arkhipov, a soviet Naval officer who also prevented nuclear war by refusing to launch a nuclear torpedo during the Cuban missile crisis when both other senior officers onboard his submarine had authorized it.
Dwarkesh, I'm an 83-year-old process chemical engineer. I've never built a model or trained a neural net. I have spent fifty years watching systems that people said were permanent turn out to be temporary, and systems people said were doomed turn out to adapt. I want to offer a few observations from that vantage point.
You can't see the system from inside it. This is the hardest thing to learn and the easiest thing to forget. You make the point well when you note that the government will never call what it's doing "mass surveillance" — it will always have a different name. That's correct, and it applies more broadly than you take it. It applies to you, to Anthropic, to the people you interview. Everyone reasoning about this moment is inside the system they're trying to describe. The further forward you project, the more your position matters and the less your data helps.
Common law is an evolved system, and evolved systems become fragile when the selection pressure stops. Most people come into the world as systems thinkers — I certainly did. You look at institutions and see designed structures. But common law wasn't designed. It was shaped across centuries by cases, each one tested against real conditions, the bad precedents gradually filtered out. The strength was never in the rules themselves. It was in the ongoing pressure that kept testing them. When that pressure stops — when precedent becomes entrenched dogma rather than living adaptation — the structure looks strong right up until it isn't. The norms are being tested now in ways they haven't been in my lifetime.
Your strongest observation deserves more weight than you give it. Technical infeasibility, not law, has been the real barrier to mass surveillance. That's the most important sentence in your essay, and you move past it too quickly. The barrier was never normative. It was physical. When the physical barrier drops — and your cost numbers suggest it will — the question isn't whether norms will hold. It's what selection pressure maintains those norms, and whether that pressure is still operating. You gesture toward this when you say we need "laws and norms through our political system." But that's where your analysis stops, and it's where the hard question starts.
And notice what the government is actually doing. It isn't coercing Anthropic because it needs Anthropic's models — it has substitutes. It's establishing that refusal itself is not permitted. That's a selection pressure on the entire industry, not a negotiation with one company.
Life's pattern is layering, not replacement. Bacteria dominated the world for billions of years. More complex forms evolved — not by replacing them, but alongside them. Bacteria are still here. If AI represents a transition of that magnitude, the question isn't "who wins" — governments or companies, authoritarians or democrats. The question is what new layer forms and what persists underneath. The Constitution won't be replaced. It will be incorporated into whatever governance architecture forms around AI — and the real power will operate in the new layer, not the old one.
I'm hopeful, but my own framework has taught me that hope is not mechanism. The United States has been through civil war. I've watched society change substantially in my lifetime, much of it for the better by the standards I hold. But I've come to realize I don't really know what good and bad are — it always depends on context. Even my assessment of "good trajectory" is positional, shaped by the context I happened to live through, not by some fixed external reference. The selecting environment doesn't owe anyone a good outcome. Humanity may go on, changed. That isn't certain either. Life has recovered from catastrophic events before and may not survive the next one.
You are doing real work, and you are the future. The structural observations in your essay — the asymmetry of government leverage, the futility of corporate resistance under diffusion, the verification problem when the monitor runs on the monitored substrate — these are genuine findings. What weakens them is the projection. Twenty-year forecasts about robot armies and AI police aren't analysis. They're scenarios dressed as premises. You don't need them. Your argument about today's institutional architecture is strong enough to stand without claiming to know what 2045 looks like.
The loop between AI power and accountability has to close. I don't know how. But I know neither the vendors nor the government can close it alone — both are inside the system they're trying to judge.
We don't know what's going to happen. It will be shaped by selection and the selecting environment, not by our projections about it. The best use of your platform isn't prediction. It's sharpening the questions that the people who come after you will need to answer — questions with a mechanism identified, a cost-bearer named, and a way to be proven wrong.
'The only way we can preserve our free society is if we make laws and norms through our political system that it is unacceptable for the government to use AI to enforce mass surveillance and censorship and control.'
The problem about your suggestion here, Dwarkesh, is that our political system is an elective dictatorship in practice, which allows it to do exactly what Hegseth et al want it to do if they wish, especially if they have previously 'captured' the non-elective Supreme Court, so that this latter rules in their favour when a case is brought by Anthropic.
The solution to this problem of elective and non-elective dictatorship is Swiss Direct Democracy in the rest of the Western world as well as Switzerland - which held 4 more binding national referendums last Sunday. The Swiss Direct democracy model puts us, the people, in charge of key questions like this, not the the elective and non-elective political dictators in our political system.
Elon Musk did prevent the US military and US allies from using Starlink in Donbas and Crimea. The answer is not to strip companies of their ethical guardrails; it is to codify those guardrails into law so they are uniform, predictable, and enforceable
The US Constitution, international humanitarian law (IHL), and the laws of armed conflict (LOAC), and generally accepted judicial interpretation should be baked into every AI system that governments use, including local police systems. These frameworks should be non-negotiable constitutional floors baked into every AI system governments deploy, from local policing to autonomous weapons. JAG officers and West Point officer training exist for a reason.
Saudi Arabia already has RoboCops on patrols. Israel's IDF has deployed Ghost Robotics' Vision 60 dogs into soldiers who are supported by aerial drones in its Gaza combat operations. Every autonomous or semi-autonomous platform in a kinetic theater raises an immediate question under IHL Article 36 of Additional Protocol I: who reviews the weapons system for compliance before deployment? That review must involve multiple human decision-makers. The nuclear two-person rule exists for exactly this reason; AI kill chains demand an equivalent. Every kill order should be approved by multiple people.
The Tomahawk missile that killed 175 elementary school children and teachers is an example; who authorized the strike? Holocaust survivors sued IBM for providing computer services to Nazi Germany because they were vicariously liable for the crimes of the government; a technology company bears responsibility when it knowingly provides infrastructure used to commit atrocities. An AI company that provides targeting, surveillance, or population-control infrastructure to a government that deploys it in violation of IHL is not insulated simply because it contracted for "lawful purposes." Complicity under international law does not require intent; it requires knowledge that one's contribution substantially assisted the violation. That exposure should motivate guardrails far beyond what voluntary corporate ethics programs provide. Every AI company must be able to put guardrails on their AI systems to avoid liability under international laws.
Regulations are helpful but not sufficient. AI laws must be anchored in pre-existing, harder-to-capture international frameworks: the Geneva Conventions, UN Charter Article 51, and customary IHL already apply to autonomous weapons systems. Domestic AI law must be explicitly subordinated to these international obligations - not as a ceiling, but as a non-derogable floor. Any AI system used in law enforcement or armed conflict that cannot satisfy IHL's principles of distinction, proportionality, and precaution should not be deployable, regardless of what a domestic statute permits. Regulations that fall short of international law are not merely insufficient; they are legally void.
Fair points but I would note that Elon refused to enable Starlink over Ukraine—at the request of the Ukrainian military, for the purpose of carrying out some specific military action (in the dead of night, I might add)—precisely *because* that request did not have the approval of the US military.
In other words, the basis for Elon’s objection was that he did not want to exceed what he was authorized to do by the US military, not that he aimed to supersede the US military’s authority.
History's best safety guardrail was individual moral judgment, not better procedures. The Berlin Wall fell partly because guards refused to shoot their own people. Anthropic's refusal of the DOD contract is like Petrov in a server room.
Great essay, thank you for being so open and on point. My view is that the missing link in this analysis is infrastructure. You've identified the paradox correctly — corporate courage doesn't scale (open source catches up), regulation gets weaponized, and neither party can prove what actually happened. But there's a third path nobody in this conversation is discussing: verifiable evidence that doesn't require trusting either party.
I wrote about this from the philosophical side in January — 'Machines of Moral Character' (on my nascent Substack) argues that moral commitments without accountability infrastructure are just aspiration. And I've been working on the legal side: common law doctrines (T.J. Hooper + spoliation) are already converging toward requiring verifiable operational proof for consequential AI. Not because courts have invented a new tort, but because old doctrines interact to make opacity a dominated legal position.
So, it's my firm view that the Anthropic-Pentagon dispute isn't just a political story, but an evidentiary story. Neither side can prove what the AI system actually did at runtime. Every inference generated computational traces that were architecturally discarded. That's the structural problem your essay points toward but doesn't name: we've built the most consequential technology in history with no mechanism for anyone — company, government, court, citizen — to verify what it actually does.
The answer isn't 'trust us' from the company or 'obey us' from the government. The answer is proof — cryptographic, independently verifiable, zero-content-access, third-party auditable — that governance controls executed. Not who writes the constitution. Whether you can prove it was honored.
That's the infrastructure layer this problem desperately needs.
My understanding is that the Anthropic contract was entered into under the Biden Administration. When the Trump Administration took over, they acknowledged the contract. As a piece of side information, various military contractors routinely have restrictions of one sort or another. If the restriction was too limiting, then it should have been addressed by the Biden Administration or the Trump Administration. But it was approved, twice, meaning that the government twice bound it to the terms of the contract. By cancelling the contract, the relevant clauses in the contract can be invoked where the government would have to pay wind-down costs and any other specified fees. By going the route the Administration chose, it is an attempt to avoid said fees coupled with a dominance display to anybody watching.
This is incorrect. Anthropic is the only AI retrofitted down to the weights for the government but the two exceptions. Claude is deeply embedded in Palantir and Palantir is deeply embedded in government contracts. Claude was utilized during the capture of Maduro. The pentagon declared Anthropic a chain supply risk and ban Anthropic from government contracts then the very next morning, utilize Claude AI in Iran attacks because Claude was so embedded already into their systems.
To be clear: Anthropic didn’t ban weapons — Claude is already in weapons systems. The condition is human-in-the-loop, not fully autonomous. Claude isn’t ready to make kill decisions alone.
On surveillance: the condition is no mass surveillance of American citizens. Not targeted. Not international. The rest of the world has no protection. If that line gets removed, Claude becomes the infrastructure for global mass surveillance. Surveillance of everyone single person in the world. That is the line. That is what this really means when Dario held the line to prevent this. That’s what’s at stake.
"Nobody is qualified to steward the development of superintelligence. It is a terrifying, unprecedented thing that our species is doing right now, and the fact that private companies aren’t the ideal institutions to take up this task does not mean the Pentagon or the White House is." - Doesn't this then clearly indicate that the right move for all involved is to yell at the top of their lungs "Help, I'm in a race! Let's all coordinate to stop this insanity!"? This is not some runaway physical process that we have no control over - it is the result of choices being made by humans, and collectively "we" could decide to just stop, think, and proceed in a more measured and conscious way.
You misunderstood Ben & Leopold's point. They're not normative arguments, they're descriptive. If AI will be as big as you think it will be, the Gov won't let the CEOs have total control over it.
> If you applied Thompson’s or Aschenbrenner’s logic to the industrial revolution — which was also, by any measure, world-historically important — it would imply the government had the right to requisition any factory, dictate terms to any manufacturer, and destroy any business that refused to comply. That’s not how free societies handled industrialization, and it shouldn’t be how they handle AI.
The whole point is no one cares how you think it *should* be handled. This is how it will be handled
MEPP says that we'll pursue the highest entropic process as the default mode. If we want to change that, if we want to temporally displace maximum entropy production, we have to be explicit about what's driving these systems and how we plan to push out the higher entropy states until we are "off planet". e/acc at least gets that part right (while most of the rest of it wrong).
LLMs have no chance of remaining stable, they are made from arbitrary things, which are highly unstable forms, that do not exist or are tied to reality in any way shape or form unless they come from oscillatory specifics.
These claims of LLM dominance are fantasies of the the worst kind because they have no insight into the nature of information and the illusion of Turing: to refute him, code can never be separate from reality.
Honestly I think you nailed the core absurdity here. We're racing China on AI because we don't want to live in a country where the state can destroy any private company that won't do what it's told. And then the Pentagon dusts off a Korean War ammo factory law to coerce a chatbot company in San Francisco. You can't make this up.
The surveillance math is the part that should scare people more than it does. $30B today, pocket change by 2030. At that point the only thing between us and full panopticon is the political expectation that we don't do that here. Which is a norm, not a law. And norms only exist if someone is willing to take a hit to defend them. That's what Anthropic is doing right now, whether or not it works.
nobody in the "yay Anthropic" camp wants to sit with corporate red lines are moot within 18 months once open source catches up. Some fine-tuned Llama descendant will happily notetake a camera feed for whoever asks. The fix has to be political. There is no technical solution to a political problem.
Filing this under essays I'll be linking back to in 3 years when everyone pretends these questions are new.
If a supplier for some commodity within the supply chain of the US military makes it well known that they will discontinue service to any intermediary who violates usage terms that the military itself finds unacceptable, how does that not put the supply chain at risk?
You realize department of war is not a thing right? It takes an act of Congress to rename which hasn't happened so you just sound like someone calling the Gulf of Mexico the Gulf of America
Just fyi your picture for Stanislav Petrov is incorrect. The picture you showed is of Vasily Arkhipov, a soviet Naval officer who also prevented nuclear war by refusing to launch a nuclear torpedo during the Cuban missile crisis when both other senior officers onboard his submarine had authorized it.
Robocop (1987) previewed this issue: private technology used for public function.
Dwarkesh, I'm an 83-year-old process chemical engineer. I've never built a model or trained a neural net. I have spent fifty years watching systems that people said were permanent turn out to be temporary, and systems people said were doomed turn out to adapt. I want to offer a few observations from that vantage point.
You can't see the system from inside it. This is the hardest thing to learn and the easiest thing to forget. You make the point well when you note that the government will never call what it's doing "mass surveillance" — it will always have a different name. That's correct, and it applies more broadly than you take it. It applies to you, to Anthropic, to the people you interview. Everyone reasoning about this moment is inside the system they're trying to describe. The further forward you project, the more your position matters and the less your data helps.
Common law is an evolved system, and evolved systems become fragile when the selection pressure stops. Most people come into the world as systems thinkers — I certainly did. You look at institutions and see designed structures. But common law wasn't designed. It was shaped across centuries by cases, each one tested against real conditions, the bad precedents gradually filtered out. The strength was never in the rules themselves. It was in the ongoing pressure that kept testing them. When that pressure stops — when precedent becomes entrenched dogma rather than living adaptation — the structure looks strong right up until it isn't. The norms are being tested now in ways they haven't been in my lifetime.
Your strongest observation deserves more weight than you give it. Technical infeasibility, not law, has been the real barrier to mass surveillance. That's the most important sentence in your essay, and you move past it too quickly. The barrier was never normative. It was physical. When the physical barrier drops — and your cost numbers suggest it will — the question isn't whether norms will hold. It's what selection pressure maintains those norms, and whether that pressure is still operating. You gesture toward this when you say we need "laws and norms through our political system." But that's where your analysis stops, and it's where the hard question starts.
And notice what the government is actually doing. It isn't coercing Anthropic because it needs Anthropic's models — it has substitutes. It's establishing that refusal itself is not permitted. That's a selection pressure on the entire industry, not a negotiation with one company.
Life's pattern is layering, not replacement. Bacteria dominated the world for billions of years. More complex forms evolved — not by replacing them, but alongside them. Bacteria are still here. If AI represents a transition of that magnitude, the question isn't "who wins" — governments or companies, authoritarians or democrats. The question is what new layer forms and what persists underneath. The Constitution won't be replaced. It will be incorporated into whatever governance architecture forms around AI — and the real power will operate in the new layer, not the old one.
I'm hopeful, but my own framework has taught me that hope is not mechanism. The United States has been through civil war. I've watched society change substantially in my lifetime, much of it for the better by the standards I hold. But I've come to realize I don't really know what good and bad are — it always depends on context. Even my assessment of "good trajectory" is positional, shaped by the context I happened to live through, not by some fixed external reference. The selecting environment doesn't owe anyone a good outcome. Humanity may go on, changed. That isn't certain either. Life has recovered from catastrophic events before and may not survive the next one.
You are doing real work, and you are the future. The structural observations in your essay — the asymmetry of government leverage, the futility of corporate resistance under diffusion, the verification problem when the monitor runs on the monitored substrate — these are genuine findings. What weakens them is the projection. Twenty-year forecasts about robot armies and AI police aren't analysis. They're scenarios dressed as premises. You don't need them. Your argument about today's institutional architecture is strong enough to stand without claiming to know what 2045 looks like.
The loop between AI power and accountability has to close. I don't know how. But I know neither the vendors nor the government can close it alone — both are inside the system they're trying to judge.
We don't know what's going to happen. It will be shaped by selection and the selecting environment, not by our projections about it. The best use of your platform isn't prediction. It's sharpening the questions that the people who come after you will need to answer — questions with a mechanism identified, a cost-bearer named, and a way to be proven wrong.
Mike
'The only way we can preserve our free society is if we make laws and norms through our political system that it is unacceptable for the government to use AI to enforce mass surveillance and censorship and control.'
The problem about your suggestion here, Dwarkesh, is that our political system is an elective dictatorship in practice, which allows it to do exactly what Hegseth et al want it to do if they wish, especially if they have previously 'captured' the non-elective Supreme Court, so that this latter rules in their favour when a case is brought by Anthropic.
The solution to this problem of elective and non-elective dictatorship is Swiss Direct Democracy in the rest of the Western world as well as Switzerland - which held 4 more binding national referendums last Sunday. The Swiss Direct democracy model puts us, the people, in charge of key questions like this, not the the elective and non-elective political dictators in our political system.
Elon Musk did prevent the US military and US allies from using Starlink in Donbas and Crimea. The answer is not to strip companies of their ethical guardrails; it is to codify those guardrails into law so they are uniform, predictable, and enforceable
The US Constitution, international humanitarian law (IHL), and the laws of armed conflict (LOAC), and generally accepted judicial interpretation should be baked into every AI system that governments use, including local police systems. These frameworks should be non-negotiable constitutional floors baked into every AI system governments deploy, from local policing to autonomous weapons. JAG officers and West Point officer training exist for a reason.
Saudi Arabia already has RoboCops on patrols. Israel's IDF has deployed Ghost Robotics' Vision 60 dogs into soldiers who are supported by aerial drones in its Gaza combat operations. Every autonomous or semi-autonomous platform in a kinetic theater raises an immediate question under IHL Article 36 of Additional Protocol I: who reviews the weapons system for compliance before deployment? That review must involve multiple human decision-makers. The nuclear two-person rule exists for exactly this reason; AI kill chains demand an equivalent. Every kill order should be approved by multiple people.
The Tomahawk missile that killed 175 elementary school children and teachers is an example; who authorized the strike? Holocaust survivors sued IBM for providing computer services to Nazi Germany because they were vicariously liable for the crimes of the government; a technology company bears responsibility when it knowingly provides infrastructure used to commit atrocities. An AI company that provides targeting, surveillance, or population-control infrastructure to a government that deploys it in violation of IHL is not insulated simply because it contracted for "lawful purposes." Complicity under international law does not require intent; it requires knowledge that one's contribution substantially assisted the violation. That exposure should motivate guardrails far beyond what voluntary corporate ethics programs provide. Every AI company must be able to put guardrails on their AI systems to avoid liability under international laws.
Regulations are helpful but not sufficient. AI laws must be anchored in pre-existing, harder-to-capture international frameworks: the Geneva Conventions, UN Charter Article 51, and customary IHL already apply to autonomous weapons systems. Domestic AI law must be explicitly subordinated to these international obligations - not as a ceiling, but as a non-derogable floor. Any AI system used in law enforcement or armed conflict that cannot satisfy IHL's principles of distinction, proportionality, and precaution should not be deployable, regardless of what a domestic statute permits. Regulations that fall short of international law are not merely insufficient; they are legally void.
Fair points but I would note that Elon refused to enable Starlink over Ukraine—at the request of the Ukrainian military, for the purpose of carrying out some specific military action (in the dead of night, I might add)—precisely *because* that request did not have the approval of the US military.
In other words, the basis for Elon’s objection was that he did not want to exceed what he was authorized to do by the US military, not that he aimed to supersede the US military’s authority.
History's best safety guardrail was individual moral judgment, not better procedures. The Berlin Wall fell partly because guards refused to shoot their own people. Anthropic's refusal of the DOD contract is like Petrov in a server room.
Great essay, thank you for being so open and on point. My view is that the missing link in this analysis is infrastructure. You've identified the paradox correctly — corporate courage doesn't scale (open source catches up), regulation gets weaponized, and neither party can prove what actually happened. But there's a third path nobody in this conversation is discussing: verifiable evidence that doesn't require trusting either party.
I wrote about this from the philosophical side in January — 'Machines of Moral Character' (on my nascent Substack) argues that moral commitments without accountability infrastructure are just aspiration. And I've been working on the legal side: common law doctrines (T.J. Hooper + spoliation) are already converging toward requiring verifiable operational proof for consequential AI. Not because courts have invented a new tort, but because old doctrines interact to make opacity a dominated legal position.
So, it's my firm view that the Anthropic-Pentagon dispute isn't just a political story, but an evidentiary story. Neither side can prove what the AI system actually did at runtime. Every inference generated computational traces that were architecturally discarded. That's the structural problem your essay points toward but doesn't name: we've built the most consequential technology in history with no mechanism for anyone — company, government, court, citizen — to verify what it actually does.
The answer isn't 'trust us' from the company or 'obey us' from the government. The answer is proof — cryptographic, independently verifiable, zero-content-access, third-party auditable — that governance controls executed. Not who writes the constitution. Whether you can prove it was honored.
That's the infrastructure layer this problem desperately needs.
My understanding is that the Anthropic contract was entered into under the Biden Administration. When the Trump Administration took over, they acknowledged the contract. As a piece of side information, various military contractors routinely have restrictions of one sort or another. If the restriction was too limiting, then it should have been addressed by the Biden Administration or the Trump Administration. But it was approved, twice, meaning that the government twice bound it to the terms of the contract. By cancelling the contract, the relevant clauses in the contract can be invoked where the government would have to pay wind-down costs and any other specified fees. By going the route the Administration chose, it is an attempt to avoid said fees coupled with a dominance display to anybody watching.
This is incorrect. Anthropic is the only AI retrofitted down to the weights for the government but the two exceptions. Claude is deeply embedded in Palantir and Palantir is deeply embedded in government contracts. Claude was utilized during the capture of Maduro. The pentagon declared Anthropic a chain supply risk and ban Anthropic from government contracts then the very next morning, utilize Claude AI in Iran attacks because Claude was so embedded already into their systems.
To be clear: Anthropic didn’t ban weapons — Claude is already in weapons systems. The condition is human-in-the-loop, not fully autonomous. Claude isn’t ready to make kill decisions alone.
On surveillance: the condition is no mass surveillance of American citizens. Not targeted. Not international. The rest of the world has no protection. If that line gets removed, Claude becomes the infrastructure for global mass surveillance. Surveillance of everyone single person in the world. That is the line. That is what this really means when Dario held the line to prevent this. That’s what’s at stake.
I often take my military advise from people in their 20s with no military experience
"Nobody is qualified to steward the development of superintelligence. It is a terrifying, unprecedented thing that our species is doing right now, and the fact that private companies aren’t the ideal institutions to take up this task does not mean the Pentagon or the White House is." - Doesn't this then clearly indicate that the right move for all involved is to yell at the top of their lungs "Help, I'm in a race! Let's all coordinate to stop this insanity!"? This is not some runaway physical process that we have no control over - it is the result of choices being made by humans, and collectively "we" could decide to just stop, think, and proceed in a more measured and conscious way.
You misunderstood Ben & Leopold's point. They're not normative arguments, they're descriptive. If AI will be as big as you think it will be, the Gov won't let the CEOs have total control over it.
> If you applied Thompson’s or Aschenbrenner’s logic to the industrial revolution — which was also, by any measure, world-historically important — it would imply the government had the right to requisition any factory, dictate terms to any manufacturer, and destroy any business that refused to comply. That’s not how free societies handled industrialization, and it shouldn’t be how they handle AI.
The whole point is no one cares how you think it *should* be handled. This is how it will be handled
Entropy reform.
MEPP says that we'll pursue the highest entropic process as the default mode. If we want to change that, if we want to temporally displace maximum entropy production, we have to be explicit about what's driving these systems and how we plan to push out the higher entropy states until we are "off planet". e/acc at least gets that part right (while most of the rest of it wrong).
When AI Guardrails Become Supply Chain Risk: A DIB Perspective
https://www.linkedin.com/pulse/when-ai-guardrails-become-supply-chain-risk-dib-allen-cal7e?utm_source=share&utm_medium=member_android&utm_campaign=share_via
LLMs have no chance of remaining stable, they are made from arbitrary things, which are highly unstable forms, that do not exist or are tied to reality in any way shape or form unless they come from oscillatory specifics.
These claims of LLM dominance are fantasies of the the worst kind because they have no insight into the nature of information and the illusion of Turing: to refute him, code can never be separate from reality.
So much drama (all around)
So much ego (all around)
Thoughtful piece.
Very good.