That's not weird at all it's the difference in most cases between products and services produced by local labor vs products and services produced by more abundant, cheaper labor elsewhere. I don't complain about $20 meals because I think inequality is bad enough.
The only thing in your list that could be cheaper without underpaying local workers are pharmaceuticals.
Labor is cheaper elsewhere, yes. But people getting paid lower salaries in other countries are still getting health care, affording rent, affording restaurant meals, etc. America has a strong problem where local salaries are high and prices far outpace them, despite the country being dependent upon things produced by salaries that are a fraction of typical salaries (underpaid farm labor, restaurant staff being paid under the table below minimum wage, meat plants employing children, technology all produced in "cheap" Asian countries where locals can afford rent and get health care, clothes produced in countries that pay pennies per hour, etc).
I am a total neophyte when it comes to LLMs, and only recently started poking around into the internals of them. The first thing that struck me was that float32 dimensions seemed very generous.
I then discovered what quantization is by reading a blog post about binary quantization. That seemed too good to be true. I asked Claude to design an analysis assessing the fidelity of 1, 2, 4, and 8 bit quantization. Claude did a good job, downloading 10,000 embeddings from a public source and computing a similarity score and correlation coefficient for each level of quantization against the float32 SoT. 1 and 2 bit quantizations were about 90% similar and 8 bit quantization was lossless given the precision Claude used to display the results. 4 bit was interesting as it was 99% similar (almost lossless) yet half the size of 8 bit. It seemed like the sweet spot.
This analysis took me all of an hour so I thought, "That's cool but is it real?" It's gratifying to see that 4 bit quantization is actually being used by professionals in this field.
4-bit quantization on newer nvidia hardware is being supported in training as well these days. I believe the gpt-oss models were trained natively in MXFP4, which is a 4-bit floating point / e2m1 (2-exponent, 1 bit mantissa, 1 bit sign).
It doesn't seem terribly common yet though. I think it is challenging to keep it stable.
mxfp4 is a block-based floating point format. The E2M1 format applies to individual values, but each 32-values block also has a shared 8-bit floating point exponent to provide scaling information about the whole block.
There's also work on ternary models that's quite interesting, because the arithmetic operations are super fast and they're extremely cache efficient. Well worth looking into if that's the sort of thing that interests you.
I do wonder where that extra acuity you get from 1% more shows up in practice.
I hate how I have basically no way to intuitively tell that because of how much of a black box the system is
Well why would Claude know any of this? Obviously it's the wrong criteria. If you have your own dataset to benchmark, created your own calibration for quantization with it. Scientifically, you wouldn't really believe in the whole process of gradient descent if you didn't think tiny differences in these values matter. So...
I think you might be answering to a different person or misunderstanding what I said but you are right that just as I don’t have an intuition for where the acuity shows up in the corpus, I don’t think Claude does either
Complexity can lead to "more is different" outcomes at higher strata. I would not say reified concepts are "made up" as they can have very real effects on both higher and lower strata.
The fallacy of reification is treating something emergent as a thing-unto-itself rather than a process or interaction born from constituents at a lower stratum. A reified thing can be recognized and changed for this reason. A mental concept needs only a change of mind to mutate, or to be destroyed.
Religion may well prove to be a reification that is destroyed once it is recognized as such. But I do believe that you cannot reduce that which is real and not real to only those things that have physical antecedents at lower strata, as we see emergent phenomena in the physical world as well.
Food prices have been subsidized for decades by farmers' pride in their work and holdings, hesitancy to make a change, and attempts to maintain a family legacy. Had those farmers sold their lands off forty years ago and invested the proceeds in the S&P 500, they would be far wealthier than they are today.
It's far more practical to adapt life to space and other planets, than it is to adapt space and other planets to human life. The problem with writing a story about artificial life exploring the universe is that it's unrelatable without humans in it.
I think humans are stuck on earth permanently because there are no other environments in the solar system we can live in as easily as earth. Traditional sci-fi explained this away using overpopulation pressures, but overpopulation is fading as a reality and plausible narrative device. And forget about interstellar travel, it's too dangerous and takes too long.
If intelligent DNA-based life is still around on earth when it starts getting too hot because of the sun's expansion and it all needs to flee, it will have continued to evolve such that it will be unrecognizable and unrelatable to us. And by then I expect artificial life to have spread to such an extent that biological life on earth is basically a historical footnote.
I actually think life in the broadest sense will spread from earth to other places in the galaxy, but it will be completely alien compared to human life.
I always appreciated the Mobile Suit Gundam approach (U.C. timeline, to be specific): humans in space largely live in O'Neill Cylinder-style space colonies arranged in constellations at Earth-Moon Lagrange points, allowing them to a) be built from materials gathered from space and b) once built, manufacture things from those materials without having to land them on Earth first. There are large settlements on the Moon, and while they are important manufacturing and research centers, they're not the primary population centers in space. Mars is, to the best of my knowledge, uninhabited. As far as the outer solar system goes, only Jupiter has a permanent human presence as the primary source of humanity's helium-3 (Gundam predates the proposal of mining the Moon for helium-3 instead).
I like this approach. It's plausible based on the assumptions made by the story, lets people in space have the benefits of mostly-normal gravity and radiation exposure (as compared to, say Mars), and keeping things local to the Earth means you don't need to be too concerned with the distances involved. Where they really lose the plot, though, is with population; Gundam claims that, in less than 100 years, billions of people--in fact, the large majority of humanity--have moved to space. I can't even begin to fathom what kind of effort would be needed to build that many space colonies, and then shuttle the people up there to populate them.
This is not practical in terms of microgravity environments. You just have to read the space shuttle manual to learn how to correct a broken toilet before you understand how poorly equipped we are to handle this challenge. If you can't fix it you're back to the Apollo style bags. The toilet has a 21 day lifetime anyways. Reading the manual further shows why. Turns out lots of holes between the inside and outside are so terrible you'd rather hold onto your waste than build a convenience. Even so turds at suborbital velocities are a terrifying thought on every level.
Any loose item in the ship is an _immediate_ choking hazard. There's no gravity. Surface tension does strange things. Even to your throat. You have to trust that every person on board is going to reliably capture and contain all debris at all times. God help you if you have to fire the engines for a short period because suddenly there's "gravity" again. You can rotate a space station but you can't go anywhere really.
Medically you can't do any imaging other than weak ultrasound. No X-Rays even. Surgery is effectively impossible without gravity anyways. Not that you'll ever get that close to the effort. It's absurd how brightly we paint this dim picture in our culture and our military style propaganda.
Until we have artificial gravity I personally think we have no business in space as a civilian effort and we're not going to get particularly far from Earth because of it.
Our trajectory is to meld ourselves with machines. Fleshy humans may he stuck on earth but our machines have no such restrictions and can be engineered for all kinds of extremes.
We will eventually figure out how to imprint our consciousness into a chip. Maybe not for another thousand years but weve been building machines our entire existence to conquer nature. We will figure it out.
If there's not a constant stream of consciousness throughout the entire process, I'm just assuming that's gonna be nothing more than a copy. What else could it be? When "you" wake up, it'll be 100% convincing either way, so I assume you can only prove it going in. I'm not great at philosophy though.
Edit: biology is pretty efficient though. We might as well just start growing new bodies/parts for people, enhancing it over time. There are already functionally immortal species on Earth.
What is a continuous stream. I was knocked unconscious for a few minutes in 2007 when I came off a bike. Am I the same person today as I was in 2005?
On the other hand dementia eats away at memories and personality. My grandmother doesn’t remember having children (who are now in their late 70s), a husband (she had two), but does remember (mostly) life as a child. She wasnt unconscious though.
(It’s basically the inevitable same end game as the rom com “50 first dates”)
As you say it’s philosophical - a bit like Triggers Broom. Or 1 minute Time Machine in a way.
I like the answer to ship of Theseus (define the ship by the keel or whatever), If you were to replace every cell in your body are you still the same person? That happens multiple times over your life. What if you were to replace some cells with non human parts - a false leg, a pig heart. When do you stop being “you”
If you replace every neuron in your brain with a synthetic neuron, one at a time, a there a point you no longer exist. Is it a continuous reduction. What if those neurons are put together elsewhere, which is you. Both? Neither? A fraction of each?
A continuous stream during the process of transfer, I mean. You would have to be fully disassembled in someway to transport you in such a manner, so it’s not really like the ship of Theseus here.
I don't think we'll ever populate a planet beyond the solar system, but I think at some point we'll leave Earth. The sun has an expiration date, after all. Maybe the space-flotilla species was the most accurate outcome after all, Mass Effect!
Don't index funds trail market changes though? I thought their allocations are reactive. In other words, the Mag 7 are being bid up by people trying to beat the market. I don't see how index funds could move prices.
I do understand how they can stabilize allocations where they are, which I think is the concern. Zombification rather than a positive feedback loop.
And anyone who thinks they can consistently predict who will be among the 4% is... mistaken. Diversification is how one manages risk when a system has a power law distribution of outcomes.
Trying to beat the market is playing a zero sum game. Someone has to lose for you to win. I understand savvy winners add information, but most winners are just lucky and it still makes me uneasy to play a zero sum game.
When you simply try to match the market, you float on the tide that mostly raises all boats and sometimes lowers them. That sits much better with me.
I have investigated taking Amtrak for a family trip to do something different. "The journey is the destination" or something like that. I was branding it "slow travel" to the family so we could use it as a sort of modern life/digital detox. I also looked into a trans-Atlantic passage on the QM2.
I'm sad to report that renting a family bedroom or two joined bedrooms on Amtrak to take a journey on say the California Zephyr didn't pencil out. It is costlier than flying (about $2000 vs $1600 at the low end for both options, resp.) Even if you account for the cost of staying two extra nights at the destination it about breaks even.
With children I don't want to risk the days of travel becoming an ordeal as opposed to hours of flight time. The "digital detox" might quickly go sideways and require hours of screentime pacifiers. Maybe when they are older.
Happily the QM2 actually made financial sense and there would be more room to move about and explore the ship.
I think rail travel makes the most sense in the Acela context the article opened with - routes between cities that take less than a day. For cross-continent travel the time savings of air travel make rail travel a harder case to argue.
The point of cross continent rail travel is not being cheaper than air at all, it is about seeing and enjoying the country and the route, there is no easier or cheaper way to do that.
- A road trip would be both more expensive (fuel, hotels and maintenance/rental), strenuous and also less safe given the number of miles to be driven.
- There is quite little to see in a cruise if not near a shore or on a plane flying at cruising altitudes well above the clouds.
While times have changed and it is lot harder for parents now, I cannot help but remember growing up the number of cross country train trips just sitting at the window with nothing but a book/magazine or conversations with passengers and it was formative life experience even when quite young. It wasn't that long ago and my generation was just as addicted to tech but we were limited to doing that only on a desktop with a modem.
---
If you want to see and show the kids to help them understand the size and complex geography and beauty of the country they will inherit despite what limited time screen distractions allow, I don't think there is any better way to do it.
> The point of cross continent rail travel is not being cheaper than air at all, it is about seeing and enjoying the country and the route, there is no easier or cheaper way to do that.
Amtrak isn't useful for that. For see the continent you need to get off the train for a few hours here and there to see something. That means flexible tickets; more trains so you don't have to spend a day in a small town with 3 hours entertainment, and enough space that you can make a last minute decision to see some little tourist trap for the fun of it knowing you can get the next train.
> A road trip would be both more expensive (fuel, hotels and maintenance/rental),
Very much it depends. If you are single Amtrak is cheaper (coach seats). a family is a lot cheaper to drive, since most of the costs are fixed for everyone. You likely own the car and so are making payments anyway. Gas is the same for 1 person or a full car. Hotels are rented per room. My last trip I needed a rental car to get to the family reunion 1 hour from the station, just the cost of a rental car would have paid for gas and hotel to drive my own car (the strenuous miles is why we took the train anyway, but it was more expensive than driving)
> There is quite little to see in a cruise if not near a shore
I've never been on that type of cruise (they exist, just not what I've been on). What I've been on the sea days were near shore taking in the beautiful scenery (you don't take an Alaska cruise for the ports, you take the cruise to watch the shore on sea days), or the ship hopes between islands at night and so you are at a port all day (though next time I think I'd get a resort and stay on one island). Beware.
Amtrak is often a great choice to get around. However there are problems and they are not to be overlooked.
It is a 12 day one way trip that includes plenty of overnight non train stays(i.e. hotels), full day sight seeing stops etc .
You can mix local car rentals to take extended side trips add much longer breaks with hop-on/hop off if you plan to do so.
There are national park themed trips specifically, or number of great regional options or other cross country journeys like LA to NOLA etc.
It will be never be perfect exact fit to your specific tastes and needs, no public transit can ever be,. However that doesn't mean it is not a great option for a traveling family vacation with sight seeing and breaks, where you can actually spend quality time with the family rather than just looking straight at the road all day, while everyone else is on the phone.
Cross country rail journeys will always be the domain of weirdo railfans (I say, having ridden many of them many times). Flying is just too economical past the first few hundred miles.
However, we live along the Surfliner route, and for weekend trips it's fantastic. It's a 1-3 hour penalty versus driving depending on which city we're going to, but the kids vastly prefer it because they're not strapped in and we can all interact.
The US should focus on medium speed rail (100-155mph). It is easier to upgrade existing track than build new high speed track. There are lots of routes that aren't worth doing for HSR but would be at slower speed.
Good example is the Amtrak Cascades which reaches 80mph. The rolling stock can reach 125 mph. High speed rail would be nice, but Portland, Seattle, Vancouver may not be big enough to support it.
I disagree. The US should focus on those routes that there is ample reason to believe there would be high demand for true high speed rail (280km/h average speed including stops). DC to Boston Via NYC for example: there is every reason to believe this could pay for itself running 8 trains per hour all day. Once we have that there are lots of other cities that can be connected and as the network grows the whole becomes more useful. East of the Mississippi the US is about as densely populated as Europe.
If you already have a route in place using that is cheaper, but often you are stuck with decisions that made sense in 1850 when trains didn't go very fast. Where you are building new track is should always be build to 350km/h standards (you run at 300km/h, but build to a higher standard just in case you need to run fast to make up time at the cost of efficiency). There are many towns with populations of 50,000 or so people that you wouldn't build new track too, but if there is existing track running slower trains make sense.
The US is physically big enough that coast to coast would probably always be a fairly niche pursuit. New York to SF, say, is about 4700km. At 300km/h, assuming no stops, that is 15 hours. Most people won't want to do that.
Seattle to SF, however, is only about 1300km, so a bit over 4 hours under ideal conditions. At that point, it's probably quicker than a plane (no need for the whole getting to and from the airport thing, or the security, or the inevitable delays).
These days, being in the flying sardine tin often beats out train travel. With more adoption, I’ll bet the price difference will be closer. The comfort factor alone means I’ll take the train over flying if it’s feasible, every time. Even coach on the northeast regional is so much nicer than flying, and you’re usually a lot closer to where you actually want to be when you get off.
Depends on where you are going - for my family vacation a sleeper for 4 is cheaper than flying by a lot (i live in a high priced air travel city, I would money driving to chicago despite the higher parking costs). However I have 5 people going and so it does't work out. (It doesn't help that amtrak dosen't suggest options like 2 rooms)
We went coach amtrak which was cheaper and more comfortable than flying. I'd do that again.
Amtrak would benefit from a coach-class sleeper, like they have in India or in Eastern Europe. They just need coach benches that convert to beds. If, for a reasonable price, you could lie flat at night behind a little curtain, like you can in e.g. Indian Railways 2nd Class, it would change the game completely. Without that, you can only travel comfortably during the day, and trips are limited to about eight hours for the non-masochist. With it, cross country would be fine. It doesn't seem that hard. Lots of other railways do it.
Oh, don't get me wrong, during the day it's the most comfortable way to travel. But the ability to lie flat, even in fare classes that aren't too fancy, like they have in India, counts for a lot when you're traveling overnight.
I've thought trains are cool ever since I was a kid reading about the Silver Streak and the Orient Express, so every now and then I look into travelling by train. Unfortunately, Amtrak is like someone was tasked with making train travel as inconvenient and expensive as possible to make the idea of state-funded rail look bad. It's so bad someone wrote a book about it, called "Derailed."
Yes. Perhaps it makes more sense for people "travelling" i.e. exploring the world where the fact that it is a nights accommodation too makes it a savings and speed is not an issue.
It was always my understanding that software careers are shorter than other technical careers, and the higher wages compensate for this. More than compensate, if you invest early.
If by FIRE you mean retire in your 50s, I don't think that's an aspiration. That should be an expectation. You might be able to work a full career in this industry, but I wouldn't plan on it.
Most people don't have the temperament for FIRE. You have to live below your means, save a double digit percentage consistently, and invest.
And you have to do it for decades. You need to be able to tough it out through the worst of times (like the dot-com bubble, financial crisis, covid, and random political chaos like tariffs.)
You have to tune out the noise and always remember that on a long enough timeline, the market only goes up. And if you think it's "different" this time, it won't be for long.
The only thing in your list that could be cheaper without underpaying local workers are pharmaceuticals.