Only in the most literal sense. In a far more accurate sense what we are talking about is land where people want to live and that does grow, more or less, with population growth. It's only a handful of cities with geography that makes desirable land fixed.
That's not that far out of line. Spouse, children, two sets of parents and two sets of siblings. You're right that it's probably not 14 on average but 5-10 easily
Go to Little Bangladesh in Queens and tell me how all those people came here. It’s mathematically impossible unless they’re all illegal (which I don’t think is true) or it’s family reunification.
Each legal immigrant who becomes a citizen is the root of a tree that can continue indefinitely. If you come from anywhere other than India or China, it takes about 10 years to get citizenship from when you get your H1B. It took us less than 10. After that you can sponsor family for green cards directly. That person only needs about 6 years before they can get citizenship and start sponsoring their own family members. So it’s easy to see how we went from 10,000 Bangladeshis in the U.S. in 1990 to 300,000+ in just 35 years.
Family sponsored immigration is also severely backlogged. Look at the dates for family categories in https://travel.state.gov/content/travel/en/legal/visa-law0/v.... I don't know how many Bangladeshis are in Queens and how they ended up there, but it's a combination of illegal, asylum, and family/employment ones.
How am I responsible for what's happening in little Bangladesh in Queens? Did you go through my history to dig up where I'm from, so that you could tailor your racist reasonse?
I used it as an example because I’m from Bangladesh and have family in Queens. There’s 65,000 Bangladeshis in Queens, about half of them on welfare. That number can’t be explained by the paltry number of H1 or EB visas granted to Bangladeshis.
Okay, but in other cases it's not like that. They bring spouses, siblings, and parents who can turn bring their own spouses, siblings, and parents. That is severely curtailed these days by country restrictions so I agree that it's not 14 but it's higher than 1.
The wait time is long from the perspective of an individual immigrant (about 17 years for sibling preference for India, for example), but not that long from the perspective of population dynamics. A single skilled immigrant can bring over a dozen family members within 25 years. That's how you end up with these huge ethnic enclaves popping up all over the country in places that had very few immigrants when Al Gore was running for President.
It's an interesting philosophic question. LLMs tend to be overly verbose and defensive in coding. It's not a ton of extra complexity but it makes the code harder to follow for a human. But if a human is not writing the code how much does that matter?
I agree that they have gotten shockingly good. It's been a long time since I've seen them do something that is objectively wrong. Once we get closer to the "too cheap to meter" cost level things will change radically again.
Have you ever tried to use an LLM to add a feature to a really nice, pre 2023 codebase? It’s incredible how much easier it is to do, how much of a difference you instantly feel
> It's not a ton of extra complexity but it makes the code harder to follow for a human.
It is. I have seen Claude chasing after endless amount of edge cases that are just irrelevant in real usage especially for the kind of users we are supporting. At some point you need to stop reasoning about all those cases because it has zero benefit.
But it IS a ton of complexity and blows out context windows. Less code good. More code bad. Both for human and agent. Somehow, still only human can make code less.
> No, if you won all the time they would use you to set odds. You'd be the single most valuable customer they have.
Good movie plot, perhaps, and some (relatively) ethical old-school brick and mortar sportsbooks work that way but the non-market online gambling apps have an established system that they all follow.
They all just use 'lazy odds' compiled algorithmically by outside third-parties like Sportradar, Kambi, Genius Sports, etc and if you show any true ability to beat these odds you just get instantly restricted from betting.
Only losers are welcome to remain on the platforms. Gotta keep things maximally extractive.
Reality contradicts you. Some academics found a flaw in odds setting which allowed them to win consistently. Instead of using them to set odds, the gambling site limited their maximum bet to $1.
Not at constant processing power, memory, training, etc.
But that’s beside the point, being arbitrarily bad at everything isn’t a problem. The diminishing returns as you apply the ceiling is problematic for self improving AI.
Yes at constant resources they are significantly better.
Diminishing returns aren't necessarily a problem if the rate of increase in resources is faster. In other words, if Gen 2 takes twice the resources but Gen 1 figured out a way to triple compute efficiency then there is no ceiling.
> Yes at constant resources they are significantly better.
There has been some fundamental advancements in LLM training and efficiency. But good luck actually finding non toy models of exactly the same size, hardware, and training one from now and other from 2001 where the newer model is dramatically better at literally everything.
Most of the real world advances are from throwing ever more resources at the problem.
> Gen 2 takes
Diminishing returns are not a question of a single generation. Gen 2, 3, 4, 5… would also need to have the same 3x return on 2x resources or you don’t have an exponential curve.
I made a building and had astra fill out the interior of the bathroom with toilets. It put 6 of them in two rows back to back with no way to reach the second row. Other than climbing over the stalls of the first row I suppose.
We aren't expected to. We (or our elected representatives) passed a law saying that experimental spending gets tax breaks. And I don't really see an argument that LLMs aren't in an experimental phase.
I'm happy to make that argument. LLMs were experimental in 2023. It's 2026 now, and enormous numbers of people use them on a daily basis (ChatGPT has 1.2 billion weekly users now, according to Sam Altman at OpenAI DevDay).
They're only experimental in the way that computers are still experimental, since we continue to find new ways to use computers.
I mean this is clearly a huge stretch. Presumably the intent of this law is to incentivize novel innovation. In this case we incentivize nothing. Data centers are commodities.
LLMs are not and Llama did pretty much exactly what this type of program encourages. It was a pilot that made no money, was thrown away, and spurred a ton of independent research.
This is also a weird split brain with HN. Half the front page is mocking companies for throwing away money on AI. That Chinese models (which surely got a huge boost from LLaMa) are going to render all this investment a gigantic loss.
No, the demand is effectively infinite at the moment and models are getting cheaper to serve. At least for a while price will go down when keeping capabilities constant.
reply