jedharris

Maximum Likelihood Estimation

Changing the laws or enforcement will take a while and software won't help directly. But extensive crowd-sourced measurement of externalities would be a big deal and software should help with that.

Actual measurement (and definition) of externalities will be very diverse, so the tracking will have to evolve and expand. Crowd-sourcing can handle that.

Once externalities can be clearly documented, public opinion will shift and companies will start to change their behavior in response. Then enforcement and laws will follow.

I'm sure the last few Fed chairs would like to know that they are guaranteed success in raising inflation! Now you just have to explain why they weren't even able to get it up to 2% even with a guarantee.

In effect we somewhat have this kind of system with Food Stamps and housing vouchers -- "currencies" that are restricted to certain types of expenditures. Also of course there are advocates for education vouchers. Etc.

Unfortunately these targeted systems have been studied and results seem to be better if we just give people unrestricted grants. Slicing and dicing the money ahead of time seems less effective than letting the recipients decide.

Great point about TIPS! Effectively everyone who's arguing that we're in for catastrophic inflation doesn't trust the wisdom of the market as much as their own reasoning based on Economics 101 models.

I wonder what that tells us?

You say that in the 1980s "most countries moved to fractional reserve banking rather than full reserve". This is at best confusing: in most developed countries fractional reserve banking, regulated by central banks, goes back to the 1600s. The bank runs that have occurred throughout US history were only possible because essentially all our banks used fractional reserve banking.

Also, the 2008 crisis was not due to or amplified by fractional reserve banking. The debt overhang was due to mortgage brokers and investment bankers (often without a retail bank license) selling securities -- effectively bonds secured by mortgages. As with any security, these had the explicit risk of becoming valueless, but this risk was hand-waved away by corrupt bond rating and other tricks.

In fact, I'm pretty sure that no bank customers lost any deposits due to the 2008 crisis. Fractional reserve regulation worked exactly as intended.

The GFC does not count. The GFC was caused by (1) bad underwriting policies (banks, AIG) that generated a huge debt overhang and (2) bad and poorly understood dependencies between banks that led to breakdown of the normal flows of assets between the banks.

None of this was in any way dependent on or catalyzed by government money creation. The same dynamic cause the beginning of the great depression in 1929-30 due to excessive margin lending and then runs on banks. At that time the US was on the gold standard.

Key observation. In recessions (liquidity crunches) the cake grows more slowly. Maybe if we increase liquidity a lot (i.e. print money) and give it to the bottom tiers who will spend it (increase velocity) then the cake will grow faster!

We don't have good models for how the economy actually works. Until 2008 most models didn't even have a financial sector. Currently economic models can't accurately predict the velocity of money or even explain changes after the fact very well. We still have no consensus among economists as to what causes recessions.

So this is an experiment but the claims that it is "obviously" going to lead to inflation are flat wrong. We are going to learn a lot from this experiment. If we need to pull back to bring down inflation, we have tools for that.

You say "There is no magic or mystery here. This is not complicated. It is not new, unknown and to be determined."

Clearly wrong. Specifically consider the velocity of money. If it falls as the supply increases there will be no inflation. It is falling <https://fred.stlouisfed.org/graph/?g=AzYM> (from another comment). Why? Will it continue to fall or will it rise? You don't know and can't figure out.

Also, how much liquidity is required to maximize economic growth -- the "size of the cake"? In recessions (liquidity crunches) growth falls. Would it rise further if we increase liquidity beyond "the usual"? Would it rise enough to wash out the inflationary effect? You don't know and can't figure out.

Your positions and other similar ones depend on the situation being subject to simple reasoning but this is wrong. The interactions are complex and the obvious variables do not provide a clean way of decomposing the system.

As Noah says in the article. The question he's addressing is how to know when we are approaching the cliff. We all agree it is out there but we can have a much better society if we know how far. If we fall off the cliff of course that is bad. Conversely if we cripple our opportunities because we are afraid, but the cliff is really many many trillions of dollars further away, that is also bad.

Recently the Fed hasn't been good at getting inflation to 2% (their published target) and economists haven't been good at predicting whether the Fed will succeed, or modeling why the Fed has failed.

Noah (article) doesn't emphasize this but he does refer to it. We don't really understand ordinary inflation much less hyper-inflation. Even the cause of the famous period of inflation in the 1970s is still very much debated.

This kind of behavior can kill the legal system pretty easily. Responding takes much more time and effort than committing the offenses in the first place, the sanctions have little effect, and the lawyer continues to practice. Probably he'll just ignore the order to send copies of the judgement to his clients.

The problem is the same with conspiracy theories -- they take way more time and effort to refute than to spin, and the proponents just ignore the refutations.

Of course these aren't distinct -- the conspiracy theories are easily converted into absurd court cases. However in the recent incidents the lawyers seem to have tried to avoid sanctions. Trump should probably have hired sovereign citizens who aren't limited by those considerations.

I think this is a structural problem with our social systems -- they depend on people at least minimally following the rules. If some groups just ignore the rules egregiously enough the system can't effectively counter. Kind of a distributed denial of service attack.

Maybe giving judges the ability to lock up or disbar people like this would work for legal problems. Unfortunately we have no equivalent likely fix for those who spin conspiracy theories -- deleting twitter and facebook accounts doesn't seem very effective.

Same topic as a year ago, but deserves much more examination than it got then.

This paper is a great overview of the state of the art in building flexible generalizable deep learning systems. Not easy to read, but if you puzzle through it you'll understand where the field is and where it is going.

This paper drives a stake through the heart of claims that explicit symbolic reasoning is needed for intelligent systems. It builds extra-symbolic (deep learning) systems that substantially outperform symbolic reasoning on tests specifically designed to require symbolic reasoning.

Since today's programming is 1000% explicitly symbolic, the potential shift to better methods that are extra-symbolic will be pretty disruptive!

The paper doesn't help to answer: Why now? The model it proposes would have produced high polarization any time, but polarization is much higher now than 40 years ago...

If the model could help us analyze the dynamics of polarization it would be helpful. As a static analysis it is just a fairly obvious application of cognitive dissonance.

The author's most recent version of KataGo averages about 50 times greater learning efficiency from self-play.

See <https://blog.janestreet.com/deep-learning-the-hardest-go-pro.... This covers developments since the initial post, plus the headline problem (which is still open).

Related demos, repositories, papers and posts:

More recent and extensive post: https://blog.janestreet.com/deep-learning-the-hardest-go-pro...

Site to play against KataGo: https://online-go.com/player/592684/

Arxiv paper on KataGo design: https://arxiv.org/abs/1902.10565

GitHub repository of KataGo: https://github.com/lightvector/KataGo

Another writer commenting on the history of Go AI, and significance of KataGo: https://brantondemoss.com/writing/kata.html

Why didn't the investigator subpoena the details about the accounts sending the accusations? That would have shown right away that the accusations were not "credible" and the university could have just dropped the investigation right away.

By the second go round, when the "y'all" message arrived, the credibility problem was obvious but the investigation didn't address it.

This sounds to me like a very limp bureaucratic response by someone just going through the motions, with no real effort to resolve matters.

Could the university investigation subpoena the information behind the email accounts? If not, why not? This seems like a basic investigative move, and unless they are not allowed to do it, makes me doubt very much that they were serious about the investigation.

Major steps beyond popular paper (https://news.ycombinator.com/item?id=14167620). Includes extensive examples (video DSL, proof assistant built using their core). Lends itself to experimenting with type system extensions and integrated DSLs -- in principle for any language with a sufficiently advanced type system.

Would this help? <https://dl.acm.org/doi/10.1145/3302516.3307352> GPU-accelerated fixpoint algorithms for faster compiler analyses.

Brute force but maybe that is what is needed.

Makes sense to me, but I would like to have specific data that would help me understand: - How much money work needs to be invested for what incremental effect? - What kinds of investments are most effective? - How much does effectiveness depend on specific cultural factors, vs. being culture independent?

And many more similar questions.

Regarding complexity: Probably anything with a brain has the problem of balancing lots of different sources of information and maintaining (enough) coherence of behavior. So even very simple worms will need a simple version of this...

Specifically, "Signatures of criticality in a maximum entropy model of the C. elegans brain during free behaviour" <http://cognet.mit.edu/proceed/10.7551/ecal_a_010> and I'm sure many more.

Hard to identify delusions in worms. But the entropy / free energy models work for the best understood worm brain (C. Elegans).

Very long and well argued review of sources of excess mortality in developed countries. Bottom line: Obesity! and other non-medical factors like automobiles. Big implications about how we should spend money to increase life expectancy.

I appreciate your sharing your experience and think it is valuable.

Those who are upvoting: Why is this interesting / important?

How much of the empirical cognitive science and neuroscience research have you read? Graziano (discussed in my reference [1]) SPECIFICALLY addresses how and why our brains generate the primordial and weird experiences you mention.

We know lots of ways our brains generate weird experiences, some built in (like dreams) and some produced by special context (any number of illusions). Why should consciousness have some special status?