kjshsh123

So this is wrong mainly for two reasons:

1. We consume a lot more now than we did before. You could work less and consume more than you did in 1973 or whatever. But you wouldn't have the newest phone, half your kitchen's appliances, seatbelts or seatwarmers and a gps in your car, or air conditioning in your home. Even average home sizes have nearly doubled since 1973.

2. There is no economic law that productivity should equal wages. When there's decreasing marginal returns to labor, it's marginal productivity that determines wages. That's the current case. When there's increasing marginal returns to labor, then you get the corner solution where wages equal average labor productivity. That was rhe case before and what changed in 1973. That's when America's frontier closed and instead of growing outwards, America started to be forced to have to do more with what it already had.

The way out is to reopen the frontier which largely requires relaxing land use laws to allow building up and out.

Keynes is 100 years outdated as an economist. He's more relevant than most for some of his contributions. But his predictions and reasons for them were clearly not all correct.

He doesn't miss these things like gaming or edge cases... He's just talking specifically about how local LLM inference is not economical and he is right.

Anyone choosing between a cloud provider vs local specifically for an LLM has likely noticed this already.

Anyone interpreting this as "Altman wants to take DRAM away from me" or "Altman doesn't understand use case X Y or Z" is just strawmanning.

The most honest interpretation of the quote in the article (if you actually read it) is he says:

1. There's a problem which is the shortage

2. People assume the solution to the problem is just everyone has local inference

3. That won't work because if everyone has local inference there's actually a worse problem because it's less efficient (idle time, partial batch sizes etc) so we'd need even more compute

There's nothing that even remotely suggests he wants to do something nefarious from this quote. If you hate him, find a better reason like the weird cornea scanning cryptocurrency ideas.

I guess Trump is going to have to stop bragging now about the US government buying Intel stock.

Everything that has happened is 100% aligned with ECON 101.

Demand has increased for high-bandwidth memory. That increases prices and quantity.

The increase in quantity has happened. Quantity sold has increased. Fabs have increased production and shifted away from other products including consumer products.

Prices increase when there is an increase in price of a substitute in production.

New fabs are planned, consistent with profit-maximizing behavior increasing production in response to increased prices, but they take time.

There are already open weight models that are quite capable and can use any provider.

I feel like you're implying that the mere act of consumption is good for an economy. It is not and the idea was critiqued 200 years ago by Bastiat in his parable of the broken window.

It's perfectly good for the economy if wealthy people save/invest in production instead of consume themselves.

Pointing out that poorer people have higher MPC simply doesn't mean that redistributive policies would increase production or consumption. It's entirely possible that such policies would decrease overall savings rate which you seem to think would be good, but would reduce total production and consumption.

No, that's not what "the experts" declared, some that have still put asterisks, and some people simply aren't experts.

Not sure what started this wave of CPI skepticism but it's so banal.

Content shouldn't target 10 to 14 year olds, sure. But, in practice and mathematically/statistically, I'm not sure what that means. Even if algorithms ignored age completely, there would be statistical correlates.

We should probably ban algorithms suggesting content and ads for dark things for everyone i.e. cigarettes, alcohol, gambling, misleading health content. It is honestly bad for everyone and that's actually the only way to prevent kids from seeing it.

It's not just about ban evasion. It's also about kids seeing this content everywhere.

Wealth and consumption follow a power law. That's not broken.

That's pretty near to as intended. Inflation doesn't mean an economy is broken.

We need to stop calling gambling gaming. And stop calling internet gambling/betting igaming.

I can go ask AI to help me learn something about coding. If it can be used as tool, it is a tool. That doesn't mean all uses of it are as a tool.

A project could be used for learning. Anyone could always cheat and have someone write it for them. They could also go to a professors office hours and get advice instead.

Now anyone can have AI write for them. They can also limit AI purely to advice if they choose.

The AI is random argument never made sense to me considering it can trivially be made deterministic.

The argument rests on defining a tool as a tool only if it's a tool for a specific task.

For example, if the task is hammering, then a hammer is a tool. A clamp is not a tool because it's not a tool for hammering. But clearly they're both tools for joining two things.

Sure, AI isn't a tool for reading the way a Kindle is. Sure, AI writing code isn't a tool for coding.

But you can use AI like a dictionary to aid normal reading. You can use AI like Stack Exchange to aid coding by hand.

AI is a very multi-purpose tool.

If you're using it to code for you, then sure it's not being used as a coding tool. But it definitely can be.

And if a game designer can now make games with it without contracting programmers, then it's a game-making tool.

Consumer demand can still be distorted.

I said this in another comment:

Imagine if we subsidized things that were an orange colour. Market prices of orange things would go down and consumption of orange things would go up. It wouldn't make sense to say "this is helping the many people that consume orange things therefore it is good".

It might be more economical for cargo to be moved by other means or for many people to live closer together than they do.

Imagine if we subsidized things that were an orange colour. Market prices of orange things would go down and consumption of orange things would go up. It wouldn't make sense to say "this is helping the many people that consume orange things therefore it is good".

Nope. If it was most profitable to quit the market or stop maintenance before LVT, it would be most profitable after.

A conversation with an LLM where you paste this exchange into the context is likely to clear up the reasons for your misconception.

LVT doesn't change the profit maximizing rent of apartments, or the quantity supplied.

You can feel free to solve the profit maximization problem of the apartment owner if you want.

https://en.wikipedia.org/wiki/Profit_maximization

>In the instance of perfect elasticity of the demand or perfect inelasticity of the supply, the price will remain the same and the entire tax burden is on producers. An example of perfect inelastic supply curve is unimproved land (the supply of improved land is elastic because more or less could be created by investment in improvements) or crude oil. Thus, the whole tax burden is on landowners and owners of the oil.

https://en.wikipedia.org/wiki/Tax_incidence

It's literally econ 101 that says landowners will bear the burden of a land value tax.

Supply of land is perfectly inelastic and land value tax is not a marginal cost of production so does not change MR=MC.

Actually no, land value tax is not a marginal cost so does not change the profit maximizing use of land.

The idea of cost-plus pricing is folk economics.

Tax incidence is very well understood in economics and has to do with relative supply and demand elasticity (supply of land is perfectly inelastic) and marginal costs which land value tax does not touch.

It's accurate to say LVT changes the price of land. But it doesn't change the profit maximizing productive use of land.

Other taxes that scale with production (sales tax, income tax, property tax) do change profit maximizing productive use.

Tell me where I argued against everything that's not LVT lol

Again strawman and false dichotomy.

I can't help if you think LVT wouldn't reduce the disparity sufficiently. I'm saying that it's possible it does. I outlined the process how. You don't accept it that's fine. But I do think a big LVT would make a heck of a difference.

Again, that doesn't mean I'm against every and any other tax that could also make a difference.

And what is a reasonable level? And what evidence do you have that LVT can't achieve that? Maybe LVT creates such an increase in productivity it creates a virtuous cycle where land values increase so LVT increases and we can pay for so many public services and subsidies for the poor.

Idk but saying it's fixed when it's reasonable is really not answering the question.

If it's also a necessity for life, then honestly, yes?

Also, even when not a necessity, taxes are often just more efficient than the alternative.

Think about taxi licenses before Uber. The government thought there were too many taxis. They made a limited supply of medallions therefore. The government gave them away to existing taxi drivers for free. Eventually they reached $1,000,000.

Eventually there were too few taxis. Existing owners complained about trying to expand the supply because maybe they just took out a loan to buy one. People were even renting out medallions. Something the government handed out for free.

The whole mess could have been avoided if the government just taxed taxis until the quantity matched what they wanted.

In a way, you'd end up in a position similar to just handing out the medallions for free. The differences are the government is getting the money and the rate can be more easily changed.

This is a pretty close metaphor for land value tax. All LVT really does is instead of paying the prior owner or a bank for land, you pay the government in installments.

I don't know what degree "fixing" means.

You can tax certain things a certain amount and you get diminishing returns. For example, if you had a 50% wealth tax, it would stop collecting anything pretty quickly and would cause pretty big economic disruption.

Land value tax has the potential to collect a lot of money with less disruption.

There's like $20 trillion worth of capitalized land value in the USA. That could mean like $1 trillion a year on-going assuming 5% discounting. Land rent is a perpetual stream of wealth.

There's only $8 trillion billionaire wealth. So already 5%/year would be less. And there's a huge incentive and ability to avoid it by moving.

And I rarely ever even hear suggestions for wealth tax rates around 5%. So then if you had a wealth tax, well land is wealth, and you'd be leaving a ton on the table as you could still have the land value tax at 5% of the capitalized value.

That money could be spent on public services that poorer people will benefit from more like transit, schools, or even local hospitals and subsidized healthcare. A typical point by LVT supporters here is that these services have the impact of raising land rents.

Most taxes have this dynamic:

tax -> invest in public services poor people use -> land rents go up -> land owners benefit

LVT has this dynamic:

Tax land -> invest in public services poor people use -> land rent goes up -> tax land more so land owners don't benefit

Landlords, developers, builders etc. should earn the return on the constructed housing, not the land.

I think land value tax is good.

That doesn't mean LVT is a panacea.

I think that's a bit of a strawman and a false dichotomy.

The use of copyrighted material is unethical, I agree, but I don't think it follows that we should consider the replacement of work with technology itself unethical.

Maybe someone can explain why RL is even needed for post training with Jev? We have supervised labels.

I guess it's due to the calibrated decision part (and that's what LLMs tell me).

But I figure some supervised classification post training would still improve the model.

Humans sometimes don't remember things 100%. LLMs can produce certain text and be run deterministically.

I don't really understand this side of the debate other than as a gotcha tbh.

If LLM use atrophies your brain and skills that's bad. If it has a repulsive writing style that's bad.

I'm not sure what the debate about whether an AI is a statistical parrot unlike humans accomplishes. Is relying 100% on a bad human speechwriter somehow better?

On a purely monetary basis it probably never will.

You're competing against companies that get tax breaks, locate themselves optimally, and have large economies of scale.

Also, if it did, the hardware would be bought up, raising the price until there was no economic profit again.

If you can find a unique application for it then maybe?

Like other comments on OP have pointed out, the term at this point is vague and used for various reasons.

But given the term's emergence in a country with a communist government, and the origin described by wikipedia, I think what I'm saying is more aligned with the term's origin.

Whatever the term means on Chinese internet sounds like standard (I'm sure partially valid) doomerism. I can accept they use it to describe that.