Straw

Adam Smith would be considered a right wing economist today, similar to Milton Friedman, perhaps with more support for certain government functions such as education. However, even on that issue, given the wildly different level of education available today, its not easy to predict what he would think.

Often quotations from him are taken wildly out of context to support a 'progressive' viewpoint. An example (From David Friedman's blog):

    "People of the same trade seldom meet together, even for merriment and diversion, but the conversation ends in a conspiracy against the public, or in some contrivance to raise prices.”

But, if we keep reading:

    It is impossible indeed to prevent such meetings, by any law which either could be executed, or would be consistent with liberty and justice. But though the law cannot hinder people of the same trade from sometimes assembling together, it ought to do nothing to facilitate such assemblies, much less to render them necessary.

Adam Smith was a proponent of free trade, free markets, and flat taxation (proportional to income). Not very socialist.

(https://daviddfriedman.substack.com/p/discovering-what-is-tr...)

See https://ukfoundations.co/

Similar argument to the article, but in much more depth.

In a nutshell, the UK has made it legally difficult to construct new housing and many forms of infrastructure- electricity plants, roads/tunnels, railway, hospitals, etc.

As a result, these things take up a larger fraction of ordinary people's budget and also limit mobility and hence productivity, resulting in poverty or effective poverty even for someone with an income that would make them globally rich.

Current models don't predict the climate overall getting drier with climate change- in fact average rainfall goes up slightly.

Some areas will get drier, others (like the Sahara and Sahale for example) have and will get wetter.

The number of correct digits in the answer grows exponentially with iteration count- "quadratic convergence" means we roughly square the error each time, so after many iterations we raise the error to an exponentially high power.

I also find there are benefits to spending (on my) education. But if we can't show benefit for a specific kind of public spending, is it justified to take other's money and spend it on "education" in a way which has no measurable impact? I don't think so.

Name calling isn't an argument.

It might be appropriate to spend a lot on education, if there were a benefit! But as you can see in the first chart, its not at all obvious that all this increased spending has had any measurable positive effect.

See https://slatestarcodex.com/2017/02/09/considerations-on-cost...

Educational spending per pupil has gone up in real terms, by about 2.5x over 40 years. This is already inflation adjusted so there isn't any comparison to the inflation rate. I'd say a 2.5x increase is pretty dramatic!

That's federal only!

Wages don't count benefits, total compensation has risen faster than wages.

The average of its errors is 0- ie it is not biased. Ofc, the average absolute error, which in English one could very reasonably refer to as "average error" is much greater than 0.

The average error is in fact 0! The average absolute error is reduced but not 0.

Only in the sense that anything is a markov chain with some sufficiently large state size. It tells you nothing about the behavior.

This paper states a completely obvious tautology which is that if you limit the context length, a LLM has a finite state size!

It's called the generalized mean:

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

I agree with your sentiment, however its not exactly true.

According to his son, Milton Friedman used to say this but stopped because he realized that trade itself is free lunch- because its a win on both sides- consumer and producer surplus. He replaced it with "always look a gift horse in the mouth".

If the RNG has < 226 bits of state, such as most non cryptographic PRNGs, you cannot reach all shuffles even if you use multiple numbers.

Was not expecting Touhou music!

This only looks at wages, not total compensation. Benefits increases have made up much of the difference.

When you look at productivity vs total compensation, there may have been a small decoupling but its far less.

This has very little to do with income or wealth inequality, and everything to do with bad local government policy leading to insanely expensive housing. Anyone on a full time service job in the Bay Area is very rich on a global standard!

What's a "living wage"? Its easy to shift the goalpost! I'd bet even those service workers who can't afford absurd Bay Area housing still have a living wage by the standards of 1900!

Japan built two Yamato-class battleships, the Yamato and the Musashi.

You can implement CLOS with macros, but not the other way round!

Seems like a big benefit would come from not forking the codebase into two versions!

[EDIT] long before deep learning!

People did GPGPU computing long before GPUs. Simply look at the list of tested, supported projects on their docs page!

My understanding from chatting with them is that tensor core operations aren't supported yet, so FlashAttention likely won't work. I think its on their to-do list though!

Nvidia actually has more and more capable matrix multiplication units, so even with a translation layer I wouldn't expect the same performance until AMD produces better ML cards.

Additionally, these kernels usually have high sensitivity to cache and smem sizes, so they might need to be retuned.

I worked for spectral compute a few years ago. Very smart and capable technical team.

At the time, not only did they target AMD (with less compatibility than they have now), but also outperformed the default LLVM ptx backend, and even NVCC, when compiling for Nvidia GPUs!

Natural gas has about 5% impurities other than methane. There's no need to change standard terminology when discussing energy.

Countries who use coal have even more emissions. 20% of US power generation comes from coal even now. More natural gas now, displacing coal, means less total emissions, even ignoring the particulate and mercury pollution from coal. So right now, natural gas reduces CO2 emissions!

It's funny how the article portrays this as a key driver of climate change, when in fact natural gas displacing coal has significantly _reduced_ US carbon emissions over the last 20 years!

Its distinct, but not very- its an EMA without assuming uniform time. The stability of EMA has nothing to do with integrators in control theory and neither do these models.

These models aren't really RNNs- they have only a linear gate which cannot depend on previous tokens at this layer, so they cant update their state in a way which depends on the current state very much.

No, all of these use the same fundamental architecture with minor tweaks, such as the dynamic gate for mamba or an outer product paramterization of the values for RWKV-v5