comnetxr

no examples of "unreasonable effectiveness" or even "effectiveness" are given; just a semi-plausible technique and some questions that might be worth answering. I hope the term doesn't get diluted with more examples like this.

also not clear why a 2D local representation is being used for 1D data. there is no meaning to the ordering of the rows (different individuals who represent the samples in the genome), so it doesn't make much sense to encode that into the image. I would presume that not much meaning comes from neighboring mutations that are separated by a long string of no mutations, so the in-row locality should be broken into chunks. Neither of these basic considerations is motivated in the text either.

I would guess there is no effectiveness of CNN at all on this data set and a different statistical technique should be used on this data set...

This seems to assume that the Court's initial resistance is a signifier that the policy was _truly_ illegitimate, and their eventual acceptance was a fake legitimacy being bestowed upon tyranny. However, there are plenty of times where the Court's initial resistance was the tyranny, and the eventual acceptance was to _remove_ legitimacy from an illegitimate policy. Besides the obvious cases where the supreme court upheld segregation, I would argue that on average the trajectory of the court has been towards properly legitimate policies under our Constitutional principles rather than towards greater and greater perversion of them. On other principles, like federal power vs states rights, there has been a clear trajectory but I wouldn't identify either side as clearly legitimate or clearly tyranny.

It's a bit complex for a comment, but electromagnetism defined in terms of differential forms generalizes nicely to all dimensions. There certainly is a 2+1D electromagnetism.

According to the model in this proposal, corporations report their "true" profits to shareholders, and artificial "profits" to tax authorities using corporate accounting tricks (i.e. https://en.wikipedia.org/wiki/Double_Irish_arrangement) to lower tax liability.

This proposal assumes that, when the tax rate is based on the value of reported shareholder profits, corporations will continue to report their "true" profits to shareholders, rather than just reporting the values returned by corporate accounting tricks to shareholders. I doubt that would be the case. The best argument for it I've seen is "Presumably they want the stock to go up so they will report a profit", i.e. investors will move to companies that are on paper more profitable. But actually, investors do their research and can update their profit models of companies; a higher reported profit number will be seen as (and will actually be) a liability to the company. Companies that forgo accounting tricks will be seen as forgoing potential profit by paying taxes that they don't need to. (The same is true today; if Google announced today that they were voluntarily giving up tax accounting tricks and paying full rate, their share prices would go down.)

The one caveat is whether the SEC enforces that the reported profits really are the true profits. But that model of enforcement (self reporting, cheaters rarely get caught and if they do settle the cases for amounts less than their profits) is silly. If we want taxes paid on true profits, we should implement tax law that precisely defines what those are and instruct the IRS to determine those exactly and send them a bill. Reporting requirements to the IRS should be set up with incentive structures that make cheating unprofitable or impossible.

While I would certainly be more happy in a world with either IRV or Condorcet method than FPTP, I think this argument against IRV is too simplistic. The situation is candidates D, R with the largest numbers of voters who prefer them first, and consensus candidate I with smaller first preference than D, R but lots of 2nd preference. Under any Condorcet method, the D and R voters can hurt their candidate by ranking I as second (later-no-harm criterion). This leads to a negative partisanship spiral, say, as D/R voters feel like they must only rank 1 candidate in order to counter the threat from the other party doing the same, and thus the consensus candidate I, which would be the Condorcet winner under voter's true preferences, does not become the winner under the actual votes. IRV does a better job of inducing everyone to rank choices.

I played a short game; I captured black's queen on black's back row with my queen, in which the black king could have taken my queen or moved out of check. Instead, black moved a different piece (invalid given that it is in check). So I captured black's king...

Agree 95%, but one point he uses to crticize Warren seems wrong:

Where Warren says "America’s big tech companies have achieved their level of dominance in part based on two strategies: using Mergers to Limit Competition,..." he argues that they achieved dominance by making good products and were already dominant when they started making major mergers, and that Warren doesn't understand this.

It sounds like he's missing the whole point here: google, amazon, and apple of course did achieve dominance by making a good product, but then they leveraged that dominance over one product to dominance over the whole tech sector with a combination of more good products and anticompetitive moves. Warren does not argue that companies that make good products shouldn't be allowed to get big - just that they should be scrutinized and not allowed to make anticompetitive moves. Indeed, all of the tech companies she has proposed breaking up would still be huge companies with "dominating" products (including control over their respective platforms - search, app store, amazon store); but hopefully the ecosystem on those platforms can flourish without competition from the platform companies.

It collapses the second narrative, but not the first. It could still be that drug money was the driving force behind the cartels, but after they grew to sufficient size, the cartels could sustain themselves otherwise.

The second narrative is kind of obviously too simple. Of course the cartels won't just dissolve, and maybe they will find ways to sustain themselves indefinitely. But maybe we shouldn't continue to feed them? Surely that doesn't help.

Thanks for the tip Jamii, and for the nice original post. This is pretty close to what I've been doing, but since I'm storing Julia values rather than numeric data I still to need to convert my data to strings to store in "foo.csv" and then parse back from strings when reading in individual values using `loadtable("foo.csv", output="foo/", colparsers=[parserfortype1, parserfortype2, ...])`. As a parser for nested Julia types is fairly complicated (structs containing dictionaries as fields which contain nested arrays as values which contain ...), it'd be nice to go directly to the binary form used by JuliaDB without converting to/from the string intermediate representation needed by the csv format. Perhaps Blob isn't quite the abstraction needed, but if such a thing exists I'd love to know about it!

Cool! Could this be used to make on-disk DataFrames that store arbitrary Julia types? Serializing/deserializing such dataframes with JLD has given me lots of trouble (both too slow and also sometimes the files get corrupted in ways that can't be debugged.)

I wanted to like Simple. But the budgeting and money tracking features were never really that useful, I needed checks, and my card number was stolen and used to purchase groceries in another state even though I had only used it for a couple of transactions. Their customer service handled the transaction but it really seemed like the card leak was probably on their end. I pulled my money out then.

I've tried both. I liked Asana a lot in the early days - it was simple and removed the friction of logging tasks. I especially liked that many tasks could be entered in the same way as an outline, just using enter to drop to the next line and tab/shift-tab to nest tasks. All the task properties I needed were visible at once and I could quickly tab through the task details and log what I needed. Eventually they added more features for teams and the features I liked became more complicated to use, so I dropped it.

I tried taskwarrior as well, thinking that being in the command line would reduce the friction for entering tasks as I spend so much time there already. However, the friction is higher for viewing and reorganizing tasks so I would fill up a todo list and then not use it.

The radius of the biggest black hole that would have evaporated by now is about a femtometer, or the size of a proton.

Reasonable sized black holes evaporate extremely slowly. A black hole the mass of the moon would take 10^45 years to evaporate, so none would have evaporated yet in the lifetime of this universe which is about 1.4*10^10 years old. See https://en.wikipedia.org/wiki/Hawking_radiation. Only black holes around 10^11 kilos or less will have evaporated by now. Couldn't think of any objects around that size to compare to, but there are some here: https://en.wikipedia.org/wiki/Orders_of_magnitude_(mass)

memoization is a common use case. If you are doing an expensive computation that gives the same answer for the the same inputs, you can build a lookup table for cases you have already encountered to avoid recomputing. you can use an object to encapsulate the inputs (or the inputs might be objects already).

Neither unitary or hermitian means that operators are there own inverse. Unitary here means that the inverse is the same as the complex transpose - which means that there is an inverse and it is easy to find. This is important as quantum circuits must be composed of unitaries, i.e. the circuit is reversable. Hermitian operators have no particular special property with respect to inverses.

They say that action at a distance is "not compatible with special relativity" but retrocausality is because it's not full retrocausality (can't send signals back in time) but just a weaker kind. I think they've missed the point on that as they've just copied the argument for why action at a distance is compatible with special relativity.

Why would any society that had lost all access to Asimovs foundation trilogy retain the knowledge to read one of these disks?

A sub or filter for each of the arxiv subcategories would make sense and would be easy to implement for papers submitted from the arxiv. As someone in physics, i would most likely want to contribute comments in my subfield and just view top papers (or most discussed) from a combined physics super category.

Often (in the scientific computing communities that I am in) large tensor contractions are done by reshaping tensors into matrices, transposing, and using matrix multiplication. The contraction time can change dramatically based on the order one contracts the tensors and the relative sizes of the tensors. It seems this just uses a large nested for loop - how does this strategy compare to using dedicated matrix multiplication algorithms?

My interpretation of this is that the muon measured radius of the proton is the correct one. This rules out the speculation that the muon measurement was accessing some sort of new physics which caused it to be 4% off. The key quote is here:

"Some researchers speculated that perhaps some previously unknown physics could make muons act differently than electrons. This would have required a revision of the standard model of particle physics, which predicts that muons and electrons should be identical in every way except for their masses — and might have pointed to the existence of yet-to-be-discovered elementary particles."

This is the reason for this precision research: it could give you the hint for where to look for new physics. This result, far from "deepening the proton-size puzzle", says that muons do not have any new physics. In other words, it's a null result.

> So, for example, all users have to take a selfie via the app so their profile can be manually verified to help boost trust and keep out spammers; users don’t have to provide their real name though, and can choose not to display photos on their profiles or blur photos unless there’s an active match. Users are also asked to rate others they have interacted with — and these ratings are fed into the matching algorithm, with the aim of surfacing “quality users” and promoting positive behaviors.

All of this sounds like a big improvement on Tinder. I'm not Muslim, but I'd try a dating site that worked like this. They should open it up but allow users to restrict to the Muslim subset if they choose.