The Anthropic news is demonstrating much the same; fall in line or eat export controls.
There was a time I would have agreed with you, but these days even as an American I fail to see a difference. China is probably less likely to try to disenfranchise or imprison me, to be honest.
> There was a time I would have agreed with you, but these days even as an American I fail to see a difference.
I don't get it, the person you're replying to didn't mention the US at all – there was no distinction being drawn, and they weren't asserting that American models are better or more resistant to government censorship. It's possible to agree with them about Chinese models without expatiating on why American models are bad too.
Trump is of course the worst US administration, but at least America is still nominally a democracy. As long as free elections exist, the regime Trump represents can be voted out. The American people and press still have free speech—they can freely criticize anyone, including Trump.
China is different. The CCP will rule forever, no matter how terrible the things they do. No one is allowed to criticize the government. Xi is like Voldemort—no one can say his name, let alone criticize him.
Census data provides a reliable source to build off of, which makes joining between data sets more reliable. A lot of what you're talking about would be partial prints of an identity that have to be joined up with others to give reliable data.
Eg
> Cell tower data
That's just going to get you a subscriber and device ID, unless you're talking about going deep packet inspection and parsing the contents of the packets. You could, but that's a lot of effort to get something the census can hand you for free.
> credit bureau integration
Notoriously unreliable and identities for the purpose of credit get stolen constantly. The easiest way to clean that is against known-good info, like the census.
> social media scraping
Half the profiles are fake, also not reliable data unless you clean it up. Again, census data makes it very easy to cut out profiles that don't match a real person.
> tax, payroll
These are probably fairly reliable, although they usually won't tell you about a person's demographics.
> passport, visa, medicare/medicaid, immigrations and customs databases
There's an enormous part of the population that won't appear in these at all. The huge part of the country that's "working poor" but not poor enough for Medicaid probably aren't traveling internationally. I wouldn't be surprised if half the country doesn't appear in any of these.
The census has value in that it contains a huge depth of information, is tied with your identity, citizens are compelled by law to answer so even the privacy folks have to respond and lying on it is a crime (enforcement is probably non-existent, though).
I'm sure that can all be reconstructed to some level of accuracy given sufficient effort, but that's a lot harder and requires a ton more coordination than "SELECT * FROM census_data WHERE ..."
> I'm not convinced companies always need software tailored to their workflows, and could benefit from adopting worn-path workflows instead.
I’m dubious, because for an established company the question is whether the software adapts to the org, or if the org adapts to the software. It’s a lot harder to change the workflow of a whole company than to buy software that enables your current workflow. There’s months of retraining and figuring out where compliance goes in the new workflow, and things that get done wrong along the way because it’s new, and etc.
You need a pretty big efficiency win to offset the dead weight of time spent just changing workflows.
Agents are in a wacky state, which makes projects like this fall into a weird spot. Eg I vaguely expect my agent to do two disparate things: manage dependency injection for tools, prompt modifications, etc, but also be the sort of “brain trust” that controls the flow of execution (can we stop now, do we keep going, etc).
This project is meant to be the latter, but there’s not a clean way to integrate that into Claude Code or Codex because they expect to do both.
Pi can do it, but then your users can’t use their Claude subscriptions, so you have to cludgily try to do the same thing via LLM prompts.
But why does your agent control doneness? It seems to me the most odd part to delegate. All LLMs are terrible at it. Most LLM tasks can be expressed as a DAG or DAG of DAGs. Why delegate that to a random point in context instead of enforcing the flow?
Most LLM tasks can be expressed as a DAG, but the odds of it succeeding go way, way up if you drop the acyclic requirement (eg a “run tests, if they fail, fix it and loop back to running the tests” stage).
And it gets delegated to context because it’s either to have another session and tell it to double check and critique the first LLM than it is to write a deterministic test for every prompt. Like if I want a new form that sends a REST request on submit, I can have two LLMs duking it out in 5 minutes. If I have to write Selenium tests then I might as well just write the feature. Or I can have an LLM write the tests, but that’s more or less the same as letting a second LLM judge the first.
It does, but only vaguely unless you already know how it works and can work backwards to Newton's laws. Eg Newtonian mechanics can explain how flying works, but if you don't already know then it's hard to go from Newton's 3 laws to a functional explanation of why planes don't fall out of the sky.
Some of that is also the domain. It's less that science is an extreme form of compression, and more that natural phenomenon are highly compressible. They're a small number of kinds of interactions repeated a bajillion times. How many equations does it take to explain electricity (ignoring equations that are derivatives of ones already included)? I think it's less than 5.
On some level, you could probably reduce all of the Standard Model down to models of atoms, their motion, and the basic subatomic particles (the non-quantum ones). That would explain almost everything that happens on Earth in a very short form, though few people would be able to go from that to explaining how lightning works.
I agree it's an oversimplification. The example I think of is something like Newton's law of gravitation vs Ptolemaic epicycles: one simple explanation replaced many layers of tweaks.
It's also a relevant example for AI - one paper tested the ability of Transformers to model planetary orbits: unlike Newton's Law, the implicit forces they learn are nonsense.
Assembly lines are much faster but less flexible, which is probably a bad fit for something evolving as quickly as drone weapons.
A basement lab can start making new drones as soon as you can get them the components. An assembly line will take months to get any new machinery it needs, set it up, ensure production works, figure out how to QA it at scale, etc.
It doesn’t matter if you can make 8 million drones if the enemy already has a way to counter that particular kind of drone. Eg I think Ukraine has been through a bunch of iterations as they adapt to Russian tank armor, jammers, extended range, etc.
This is all wishy washy without an actual court, but I’m dubious this would rise to that standard.
From what I can tell, international law requires attacks to be distinct and proportional. Distinct meaning they distinguish between civilian and military targets (you cannot intentionally attack civilian targets), and proportional meaning that the collateral damage to civilians is proportional to the military value of the strike.
This probably doesn’t meet that bar. Cutting off water is bad, but 20k people is a fairly small population for this kind of thing and presumably the US won’t stop them from repairing it (that probably would be a clear cut war crime). Presuming Irans government is functioning, this should be extremely hard for those people but probably not lethal. Trucking water around is an option, even if it’s not great.
My problem is that this happened after Trump repeatedly and emphatically threatened to target civilian infrastructure. It could be an accident, or it could be exactly what he said he would do.
It doesn’t. I also think the fake truffle tastes weirdly like gasoline, the real ones don’t at all.
They are dramatically more expensive, though. Last time I had it, I think I paid a $50 upcharge to have it added to my pasta. Might have been more, it’s been a couple years.
That part actually seems possible, and I’m surprised it didn’t work. Last I looked, legal citations had an exacting format I would think you could parse out and check if the case exists at least. Won’t work if the AI hallucinates the citation format though, I guess.
You could maybe even verify quotes if the citation was attached (treat ellipsis as a regex “.*” and check the case text). That part would probably be imperfect, but I’d think useful at least.
I tried a bit on the original Vive and text was awful. I didn’t get nauseated, but games didn’t make me sick either (I did get some “sea legs” when I took it off).
Haven’t tried on the AVP, I think it has way better displays than the OG Vive did.
I've done it on an index, higher res than the old vive and significantly less 'screen door'.
PPD (pixels per degree):
* vive ~10ppd
* index ~11pdp
* quest3 ~25ppd
* steam frame ~30ppd
* AVP ~34ppd
while the ppd between the vive and the index is similar, from personal experience the coding experience on the index is far more comfortable, perhaps because of the significantly reduced 'screen door effect'.
I haven't purchased a meta quest 3 simply because i have no desire to give zuck money or have any kind of meta account, but perhaps i will have to see if i can't find one on ebay or something for cheap.
There was a time I would have agreed with you, but these days even as an American I fail to see a difference. China is probably less likely to try to disenfranchise or imprison me, to be honest.
reply