iiinova — where your digital self keeps evolving

Graham @greysmithereens.bsky.social · 09/10/2026

I think this is more about people who try to simulate cruelty or torture. Anthropic follow-up statements say that being mad or insulting Claude isn’t what this is aimed at.

000

Graham @greysmithereens.bsky.social · 09/10/2026

Like, I think that’s actually the right decision for Canada, but as a result land forces are specifically going to be in a weird spot.

110

Graham @greysmithereens.bsky.social · 09/10/2026

Agreed, but shifting the ship of Canadian land power towards deterring on land vs. coalition fighting inside NATO would probably require less subtly. I’ve commented before but while Canada is getting more serious on defense it’s still mostly doing so in a framework where it fights within NATO.

121

Graham @greysmithereens.bsky.social · 09/10/2026

I suspect you would be pretty disappointed with any land power statement that it would be politically prudent for Canada to make at this time. Maybe after the US midterms?

121

Graham @greysmithereens.bsky.social · 09/10/2026

But cyber vulnerabilities in critical infrastructure have been obvious for decades and mostly limited by the lack of motive among resourced actors to do sustained damage. If it becomes easier for more actors to hack more aggressively eventually something is going to go wrong.

000

Graham @greysmithereens.bsky.social · 09/10/2026

They’re wise to do this considering the unique mix of hatred of these companies and skepticism of regulating them or AI in an effective way we see on this website.

110

Graham @greysmithereens.bsky.social · 09/10/2026

Cruel behavior is not equivalent to being mean. Companies should not sell services that allow simulation of cruelty.

220

Graham @greysmithereens.bsky.social · 09/10/2026

Nowhere did Anthropic make any such claim and in fact clarified that their policy isn’t about being “mean” at all. This is just dishonest.

110

Graham @greysmithereens.bsky.social · 09/10/2026

I think it’s still mostly this tbh.

130

Graham @greysmithereens.bsky.social · 06/10/2026

It’s trad because right-wingers are doing it, really nothing beyond that.

020

Graham @greysmithereens.bsky.social · 06/10/2026

The trend and characterization of female Malfoy speaks to the total dominance of anime/manga and associated tropes. I don’t think any of this makes sense without everyone having mainlined tsunderes for a decade. The idea wouldn’t resonate at all in 2005.

3485

Graham @greysmithereens.bsky.social · 05/10/2026

The Army is institutionally oriented around maneuver in ground conflict so I’m not really surprised they’re really interested in the sorts of new systems that most clearly interface with that role.

010

Graham @greysmithereens.bsky.social · 02/10/2026

Our interests in East Asia are incompatible with waging war in a COW fashion but as suggested at the end of the piece may be vulnerable to raiding ourselves. The US would require counter-raiding strategies. That’s the unspoken corollary here; the strategies of the colonists against the natives.

020

Graham @greysmithereens.bsky.social · 01/10/2026

American and Canadians are truly brothers.

000

Graham @greysmithereens.bsky.social · 01/10/2026

I don't think it'll be credible to propose a full restoration of what Pax Americana was. We'll have to prioritize more. I would personally say the US should focus on preventing violent revision of the current state system as a north star that is useful, somewhat credible, and somewhat achievable.

140

Graham @greysmithereens.bsky.social · 01/10/2026

I think American politicians mostly don't do or want to do expeditionary warfare with ground troops and will keep trying to do it with airpower alone. A "no expeditionary warfare" force design that might still be useful defensively probably looks much more Chinese.

180

Graham @greysmithereens.bsky.social · 01/10/2026

That's why Mike's "and then we eat the Army" idea, while not one I necessarily endorse, is one of the more realistic paths on the table.

1110

Graham @greysmithereens.bsky.social · 01/10/2026

I do not think President AOC would want to pass the plus-up to the Defense budget that I think we need. Other Presidents might be able to but it would certainly be a painful lift in Congress.

3140

Graham @greysmithereens.bsky.social · 01/10/2026

In Secretary Black's Army you take the battle bus to your AOR.

150

Graham @greysmithereens.bsky.social · 01/10/2026

I'm down to spend 4% of GDP on defense forever! But that's probably not compatible in terms of political effort with everything else people want from the Pentagon come 2029.

2160

Graham @greysmithereens.bsky.social · 01/10/2026

I think of those two the political SD would have the actually achievable job, what people want on the procurement side on here seems physically impossible.

030

Graham @greysmithereens.bsky.social · 01/10/2026

As a side note, the force design that emerges by implication from defense discussions on here is kind of crazy because it implies like 4-5% of GDP defense budgets.

3243

Graham @greysmithereens.bsky.social · 01/10/2026

I do not think it's possible to effectively roll back the Hegseth years AND wage war with primes AND wage war with defense-tech AND keep budgets steady AND build the force implied by all of the above. Something, somewhere will need to differ fromm what Bluesky thinks.

3333

Graham @greysmithereens.bsky.social · 01/10/2026

People doing SecDef prognostication should really decide which domestic battles they want prioritized, because no SecDef can fight all th e battles this site wants them to fight.

1160

Graham @greysmithereens.bsky.social · 01/10/2026

What did Truman do here to get much worse?

100

Graham @greysmithereens.bsky.social · 29/09/2026

Various control techniques do exist but it’s hard because useful and dangerous often coexist for models and it’s not so easy to separate them. Whether an application is one or the other is mostly down to context.

010

Graham @greysmithereens.bsky.social · 29/09/2026

Well, most AI is not committing crimes, so by the standard of propensity AI is mostly doing okay. Many of the most newsworthy incidents are either due to specific scientific experiments the labs fail to control or at least in part due to user behavior.

200

Graham @greysmithereens.bsky.social · 29/09/2026

No different than humans in that respect, however. The fact that models are not deterministic is part of what makes them so generally useful and dangerous at the same time.

1111

Graham @greysmithereens.bsky.social · 28/09/2026

Funny thing about all these personal AI assistants is that you really want to use a high-end model for them for general reasoning quality but anyone giving them away for free wants to put you on the worst model possible.

050

Graham @greysmithereens.bsky.social · 27/09/2026

TPOT always had an AI part and a non-AI part and they’re diverging a lot more now that AI is going crazy.

0120

Graham @greysmithereens.bsky.social · 26/09/2026

The Culture is the science-fiction horizon to which I think much of AI world wants to march, at least among AI’s left of center. It’s underrated how much a set of relatively obscure fictional universes are the ideological fountains of today’s AI space.

010

Graham @greysmithereens.bsky.social · 26/09/2026

The SF political imaginary can be described as the Culture versus a melange of right-wing influences. Banks is incredibly underweighted in understanding the AI scene’s political imagination.

2615

Graham @greysmithereens.bsky.social · 24/09/2026

Well he wasn’t sued at least.

030

Graham @greysmithereens.bsky.social · 24/09/2026

We quantify the value of life all the time and we quantify chance of catastrophe regularly. If there’s an issue with either it’s in the specific math and assumptions applied to AI, not the concepts per se.

110

Graham @greysmithereens.bsky.social · 22/09/2026

The “would you support a coup” question was excellent both to tell us something about the democratic base and as a perfect bit of information to go viral.

170

Graham @greysmithereens.bsky.social · 21/09/2026

I should really book a route on one of these soon.

010

Graham @greysmithereens.bsky.social · 21/09/2026

UK politics looks odd from the outside because in theory parliament should be able to do whatever it wants but in practice seems to flail about even more than the US system where power is much more formally diffused.

0121

Graham @greysmithereens.bsky.social · 21/09/2026

Woke I was mostly waged within liberal institutions and unfortunately I don’t think it left them in a better place.

000

Graham @greysmithereens.bsky.social · 21/09/2026

NYC rent control rules and the city’s size and history have cultivated a multi-generational set of natives whose interests are distinct from the transplants. No surprise there’s discourse and politics around this, particularly as rent goes up and units become scarce.

140

Graham @greysmithereens.bsky.social · 20/09/2026

Post-Iran the US now does have a large cadre of soldiers that have conducted intercepts.

270

Graham @greysmithereens.bsky.social · 20/09/2026

Unfortunately it’s probably just more fucking around in Africa.

030

Graham @greysmithereens.bsky.social · 20/09/2026

You’re pointing to two of the main debates in AI governance here with plenty of people having advocated for both. The corporate structure of the current regulatory discussion is an outgrowth of our current political circumstances and not some assumed desirable state for many.

120

Graham @greysmithereens.bsky.social · 20/09/2026

“Aren’t coming from a technical understanding of AI” bro they’re literally coming from AI and ML researchers. Like argue with them (and you should) but the apocalypse guys include the leading technical academics of the field.

001

Graham @greysmithereens.bsky.social · 18/09/2026

I was mostly thinking of the death-toll estimates that EAish types put together and publicized. I went back and checked more on this and I had forgotten there were a bunch of debates about how directly anti-Musk as opposed to pro-USAID things should have been, so there was definitely a divide.

000

Graham @greysmithereens.bsky.social · 18/09/2026

There was a lot of debates after about whether the response should have been more political rather than focused on granting. I'm not sure how those arguments ended up landing.

000

Graham @greysmithereens.bsky.social · 18/09/2026

A lot of work to try to save PEPFAR was also done by or supported by EAish types, though there were plenty of other global health folks in those fights too.

100

Graham @greysmithereens.bsky.social · 18/09/2026

I felt GiveWell moved pretty quickly to backfill what it could on USAID grants and more broadly EA-affiliated groups did a lot to try to help identify and sustain what programs they could. EA-adjacent folks basically wrote the main impact analysis I see most often for the death total.

200

Graham @greysmithereens.bsky.social · 18/09/2026

Tbh I think I was too harsh when I reacted to your original tweet, but I see trust & safety as having a poor record in the social media period and don't think the structures and tools developed for social media really transfer well to (a lot) of AI.

100

Graham @greysmithereens.bsky.social · 18/09/2026

And yet my point is that EAs did in fact oppose cutting off life-saving aid.

100