HN user

matthewdgreen

12,802 karma
Posts0
Comments3,131
View on HN
No posts found.

They can't do this at the same margins. Cloud compute traditionally has <= 30% margins, and that's the very generous profit margin that folks like AWS are able to squeeze out of it with lock-in and proprietary tooling (something that will probably go away as software becomes more commoditized.) Frontier model companies (e.g., SpaceX) have been promising profit margins in the 75%+ range.

Subagents are a godsend. Frontier models for initial planning and spec writing, cheaper models for implementing from the spec, frontier models for review. Frontier models to help with research. The availability of cheaper models is great, since it gives more price points for different tasks.

A lot of this is downstream of compensation schemes that explicitly reward dumb metrics, like raw paper-count without subjective evaluation of contribution or quality. I don't want to generalize, but this seems to be more common in countries that are not the US or Europe.

It's a little more complicated than that. If all you're interested in disovery, you can wander off into rabbit holes or disappear into areas that are only interesting to you. Publishability is useful because it gives you useful external feedback into whether a community thinks what you're doing is "worth discovering". Like any metric, you can find yourself gaming it (intentionally or unintentionally) and it has a slew of other failure modes. Still, you can prize discovery and also care about what other people think.

Grok 4.5 13 days ago

Wouldn't be bad if the product hadn't been paid for in advance. If it can't be delivered in working form for this car, that's not the end of the world. Tesla can just refund the undelivered product, with interest.

Grok 4.5 14 days ago

I paid for a car that drives itself in 2018 and I haven’t received it. Any updates?

I know a number of people who have gone down this route, including Senators. For example, here's Senator Wyden's proposal to add $5 billion in mandatory funding to investigate and target sexual abusers [1]. The problem with these efforts is that they're expensive: fighting child exploitation requires enormous amounts of funding.

Guess what doesn't require billions of dollars? Mandatory scanning paid for by tech companies, followed by dumping the billions of hits they produce [2] on overworked police and clearinghouses that mostly ignore them.

[1] https://www.wyden.senate.gov/news/press-releases/wyden-eshoo... [2] https://www.hstoday.us/subject-matter-areas/cybersecurity/su...

We don't need top-end frontier models to write simple applications. Opus works very well for that and it's cheaper. We need them to write things that are at the frontier.

I'm working on cryptography, all from academic research papers. Started well, but it eventually got some word into its context that is on the banlist. I found that if you tell it to fire off clean Fable subagents and you instruct it to check the Claude Code billing data to check for downgrades, you can get most high-sensitivity spec/review tasks done with Fable. Most.

I figure that once GPT 5.6 comes out, Anthropic will become interested in making the safety gate non-destructive.

I didn't understand what these satellites were really like until I visited Zion National Park two weeks ago. Zion is an International Dark Sky park, and so I was really looking forward to seeing the stars. Instead we sat outside and watched dozens and dozens of fast-moving stars zip around on all sorts of trajectories. I'm not saying it ruined the experience (I'm not an astronomer, and it was kind of fun.) But it really brought home how fundamentally we've changed the sky. I also hope we're able to lay enough fiber in developing countries that this many satellites don't need to stay up there forever.

There's a lot of stuff in this post I agree with. There's also a lot of stuff that hints that the author doesn't believe in the worst case they name: that there might be a classical cryptanalytic attack on ML-KEM that occurs before Q-day.

Sure, I already did explain it. They explicitly said that "not under the jurisdiction" would cover children belonging to Indian tribes, the children of ambassadors, and they discussed the need to be able to expel occupying armies. Someone said "what about the kids of Chinese people" and the sponsors said "yep, if we adopt this language they'll get to be citizens too".

None of it's complicated. You could read this as an 8th grader and have no doubt what they were trying to do.

Most of the discussion was of the form "hey, could we add an exception to exclude even more people from citizenship" and then the sponsors would say something like, "yes, we agree that those people aren't excluded under the current language; that adding extra language could exclude those people; and that we don't want to exclude those people or change the language." And then Congress voted for the language exactly as it was originally proposed.

What are the confusing words? You gave a quote that sums up the long-accepted understanding of who's exempt from the citizenship clause:

"This will not, of course, include persons born in the United States who are foreigners, aliens, who belong to the families of embassadors or foreign ministers accredited to the government of the United States, but will include every other class of persons."

They don't mention Indian tribes in that quote, but they mention it elsewhere. Seems completely clear to me.

The thread was about someone misunderstanding what "subject to the jurisdiction" meant. Someone suggested that it meant subject to US laws including conscription and income tax, another person said "what if the law in question doesn't mandate conscription of foreigners" and then Telemakhos gave a pretty confusing answer to that. At some point you dropped in to make a further confusing and argumentative comment in response to that, so I tried to clear it up. If you were just disagreeing with Telemakhos, then I apologize for misunderstanding.

I hope that's an accurate summary?

Here's part of the Senate debate where they discuss it. It turns out that they were extremely precise about what it meant, and they described the exceptions in great detail. It's even typewritten. The downside: you can't come away from reading this stuff and pretend like there's some legal flexibility in the term. If that's important to you, you probably shouldn't click the link.

[1] https://digital.library.unt.edu/ark:/67531/metadc30867/m1/12...

I think this take is overly complex and silly. If you're subject to the laws, that's that. As an exercise, go read the Federal code and put each law that might potentially be applicable to people on a work visa into a pile. I guarantee that pile will be taller than you are by the time you're done. It will certainly be larger than two random examples.

You're trying to impute complexity to a thing in order to achieve a goal that is not achievable. The 1866 Congress that debated the amendment understood and intended that Indian tribal nations would not be covered by the clause because they were separate nations not under the jurisdiction of US law. Here's an example of the debate [1] where they discuss it. Far from making your point, examples like this make it obvious that everyone involved in framing the amendment thought deeply about what "jurisdiction" meant, which is why you can't just assign new meaning like "yeah it means parents have to be citizens."

I really urge you to read the original debate. It isn't like the handwritten notes we get from the 1700s; it's typewritten and the Senators are so thoughtful and utterly precise about what they meant. It's ELI12.

[1] https://digital.library.unt.edu/ark:/67531/metadc30867/m1/12...

The exception listed in Wong Kim Ark are: Indian tribes, children of diplomats, births on public foreign ships, and children of enemy occupiers. These were basically well-understood exceptions under common law and most were discussed during Congressional debate on the 14th.

It's worse than that. The text of the statute is literally the same as the text of the 14th amendment's birthright clause. So he's basically saying the text means one thing when it's a statute, but the exact same words might allow exceptions in the context of the Constitution.

The great thing is that many of the Chinese open-weight models can be run on US providers. Some promise not to retain data for training or other purposes. (Which I think is a credible promise most of the time.)

I think that’s an assumption. Just because an LLM might be able to find some bugs does not mean every attacker has a packaged attack, or the right prompting. The easiest way to find a vuln is just to Google for it.