Graham

Hot take is that this has historically been the right move and “no first use” would have been wrong for the US to adopt.

There is no easy equivalent to switching AI weights mid-session but retaining that session's memories in human cognition. Debates about whether AI personhood need to account for how AI's are not anything like human people in these ways.

You can selectively edit the "memory" of AI, stop and store it for long periods of time, switch out "components" of the AI's software in ways that have no coherent human analogue, and a 100 other things.

All debates about AI consciousness must also play out against the backdrop that AI has disaggregated and made manipulable many parts of "thinking" that in biological entities cannot be directly controlled or manipulated separately.

I don't have good answers for the issue that plague defense-government contracting right now because for many of the answers I like in theory, such as greater government direction of the DiB, the US doesn't have appropriate political conditions for them to work in practice.

I worry about the political economy of any nationalization in today's era. The defense-industrial base is already much less efficient than it could be due to the congressional appropriations process, and NASA has not covered itself in glory in recent years as an example of a government-run business.

Foreign policy in the US also under multiple different sets of constraints that have seen left-liberal foreign policy thinkers' dreams dashed against the rocks even when they've had influence.

Agreed, but there's ways to mitigate that through disclosure and disinvestment, and it's not like think tankers and academics' personal economic situation isn't also implicated by decisions they make if they're in office. So long as we live in a mixed economy we'll live with this tension.

Private sector time really should not be seen as corruption on its face for a potential appointee IMO.

If anything, clerics were the primary intellectual class of the Middle Ages. It’s an interesting back-projection of today’s science/faith divide.

I would expect a blockade situation to move to war if the US chose to contest. If Xi goes for a blockade he's going to need to be ready for it to go all the way. That said, I expect the actual Plan A the CCP to look more like being invited in under a pan-blue coalition.

The problem is nobody really takes transnational response seriously. Nobody defends sovereignty more assiduously than the left of center.

Also hilarious that they're going to both designate Anthropic a supply chain risk and try to compel it with the DPA at the same time.

Really raises the risk for anyone to go into business with DoD after this if the result is that you're eventually going to get bent over a barrel because you won't revise your contract terms. I feel for Anthropic as they leaned into classified work early and are getting punished for it.

Only a difference of degree rather than kind from the old pinups on bomber noses, but I still feel like I'm having a stroke when I see anime waifus plastered on military hardware.

Apparently it's DARPA's OPEN, but I don't see how OPEN has overcomes the inherent limitations of the critical mineral market here to make investment "actually reliable" as promised.

Probably? There are definitely AI workloads that don't require good latency that could be prioritized to theoretical space-based data centers.

The headline doesn't give me any sense that this is even an LLM application. Could be just the same system that DARPA has experimented with in the past.

In the hypothetical world where you're hosting AI infrastructure in space, you're probably doing your inference up there and your training at some ground-based data centers.

Well, at the risk of being too provocative, doesn’t this suggest investment in WMDs for deterrence since conventional overmatch in this hypothetical is so large?

Given the likely contingencies you’re think of I think artillery acquisition makes sense, I just wouldn’t advise it if deterring the US was the sole priority for Canadian defense policy.

The Latvia/NATO contingencies are where this makes sense. Is your view arty helps deal with irregular separatists too? Those masks sense to me but I doubt it has much use against a true invasion by US forces.

What use are you imagining for artillery pieces here? I don't see them really lining up with the security concerns you've focused on in the rest of your posting.

All I can say is that nobody in this admin appears to believe in the Davidson Window because we’re burning readiness like crazy for all these attacks.

I've found AI-enhanced literature reviews to be useful for this, but you do need to do the reading yourself at some point.

"I'm a social scientist with a focus on inequality with no knowledge of military operations, let me tell you why Iranian hypersonics mean US carrier forces are doomed."

Depends on what type of product. I don’t see literature or persuasive writing being displaced, but much of writing is functional and llms are perfectly good at outputting functional analysis.

Had its own intelligence service! Still has counterintelligence staff! It’s a truly fascinating organization.

Much of this is wrapped up in debates about the field itself and the limits of historical evidence. Opponents of counterfactual reasoning in history will usually suggest that the sort of evidence available can't support analysis of counterfactuals, but I agree they still implicitly accept them.

Lotta folks are taking “woke II” as inevitable and seeking to relitigate the failures of woke I or bring back every idea from it. This is bad because woke I failed and whatever comes next needs to tsk those failures into account.