mjn

As relevant to the linked post (written by someone who later died of ALS [1]), there are a lot more mainstream medical research organizations who treat "death as a problem to solve". Treating deaths caused by ALS, cancer, heart disease, strokes, covid, etc. as a problem to solve is the focus of a huge amount of research.

[1] https://en.wikipedia.org/wiki/Amyotrophic_lateral_sclerosis

This is one reason I hope it continues to be socially acceptable in the U.S. to wear masks. I've wanted to wear one even pre-covid for pollution and pollen, but you would get very weird looks, or even stopped by security in some cases.

> good work gets the attention it deserves

I'm not sure I'd go that far with the optimism, at least in my field (artificial intelligence). Some obviously bad work does immediately fade into obscurity, and better work probably does on average get more attention, but the variation is huge. There is so much stuff on arXiv, that to get attention you need some kind of PR push so people notice it in the firehose, or a dice-roll around a viral tweet or science journalist noticing it. Some of the better funded university and corporate research groups have actual professional PR and science-comm teams doing coordinated social-media blitzes, press releases, and blog posts around new arXiv papers! That's a huge factor in determining whether a given paper gets attention.

> The whole point of journals is peer review. That's what distinguishes them from random blog posts. So why can't reviewers just start working independently?

It's happening slowly in some fields. Well, not quite independently, but in loose professional associations. It requires pretty coordinated action and still has a chance of failing though. One early successful case is that almost the entire editorial board of the Springer journal Machine Learning resigned in 2001, and threw their weight behind a community-run, non-profit, open-access alternative, the Journal of Machine Learning Research (JMLR), which is now considered by most people to be at least as prestigious as the Springer journal.

Open letter with their rationale from the time: http://www.ai.mit.edu/projects/jmlr/statement.html

IEEE 754 doesn't really have anything to say about which semantics you should choose for int/int division. If you choose the semantics of casting both operands to floating point and then dividing, IEEE 754 specifies both 1) how the int->float conversion should happen, and 2) how the float/float division should happen. But if you choose the semantics that int/int division returns an int without any floating-point conversions (like C's '/' operator), then no floating point numbers enter into the question at all, and it's out of scope for IEEE 754.

Same here. I won't spend money anywhere where it seems likely that I'll have to encounter unvaccinated people, especially without masks. They can infect each other if they want, but leave me out.

In particular for larger amounts of money. Many countries have deposit guarantees for smaller amounts, so for most individuals it wouldn't make any sense to buy negative-interest bonds. For example, someone in the U.S. who wants to store $250k with a government guarantee can simply put it in an FDIC-insured bank account. I believe Germany is similar, though with a lower €100k guarantee per account. You can go a little higher by having multiple accounts at different banks (at least for FDIC insurance, the first $250k at each separate bank is covered). But if you have $10m or $100m to store, government bonds become the only practical way to get a similar level of insurance.

That was unfortunately short-lived. ACM announced on March 30, 2020 that they would open their DL for 90 days due to the pandemic [1]. I don't believe there was an extension, so it expired on June 30, 2020.

[1] https://www.acm.org/articles/bulletins/2020/march/dl-access-...

It's simple compared to modern machine learning, but linear programming [1] was initially developed by Leonid Kantorovich partly with the goal of automating central planning. Most of his ideas weren't implemented in practice though.

Here's a recent paper surveying some of that work and how one might approach the central allocation problem with the benefit of modern techniques: https://arxiv.org/abs/2005.01539

[1] https://en.wikipedia.org/wiki/Linear_programming

My first paper as an undergrad measured MIDI round-trip by tapping the input and output MIDI wires and then looking at how the signals were offset: https://www.kmjn.org/publications/MIDI_ICMC04-abstract.html

For keyboard input, one paper (whose citation I can't seem to find) used a clicky mechanical keyboard, pointed a mic at the computer, and measured the offset between each corresponding pair of keyboard click and sound output.

Interesting, I had missed that. The language allowing remote searches when computers "have been damaged without authorization and are located in five or more districts" seems to be recent, added by Congress in a 2016 amendment: https://www.justice.gov/archives/opa/blog/rule-41-changes-en...

I think this is the relevant court record: https://www.justice.gov/opa/press-release/file/1386631/downl...

The mechanism seems to be a search warrant. The FBI applied for a warrant to "search" all compromised Exchange servers in the United States, and to "seize" the illicit malware on those servers by executing a specified series of commands.

A few excerpts from the above link:

"FBI personnel now seek authorization to search the compromised Microsoft Exchange Servers and uninstall the web shells on those servers". (6th page of the PDF)

"This warrant authorizes the United States to seize and copy from Microsoft Exchange Servers located in the United States the web shells identified in Attachment A, and to delete the web shells from those servers." (11th page)

That's a good point; this system certainly didn't come from nowhere! The protein datasets they used also mostly came out of various NIH-funded projects.

What I meant to focus on was that I think DeepMind has less of a pure money/scale advantage in this area than in some others. In something like Go or Atari game-playing, there are many academic groups researching similar things, but their resources are laughably small compared to what DeepMind threw at it. So you might argue that they got good results there in part because they directed 1000x the personnel and compute at the problem compared to what any academic group could afford. In biomed though, their peers in academia and industry are also pretty well-funded.

Even if you try to account for the overall R&D cost, DeepMind isn't that large an organization by the standards of biomedical research. It's very big and well funded for a computer science research organization, yes, and most CS departments can't match its resources. But the NIH budget is $40 billion, and private pharmaceutical companies do another $80 billion in annual R&D. It's interesting that this kind of breakthrough didn't come from those sectors.

As a current example: the highest-scoring entry that has an associated notebook at the moment is basically a clean example of how to apply XGBoost [1] to this dataset. XGBoost ends up being tried in nearly every Kaggle competition, so the person isn't giving away many secrets there.

[1] https://xgboost.readthedocs.io/en/latest/

Seems to depend on the PDF viewer (font issues?). The kerning is almost unreadably bad in macOS's PDF viewer (Preview), but looks perfectly fine in Firefox's built-in PDF viewer on the same machine.

> The last stable release of Cairo was in October 2018

Looking at the release history, as far as I can tell, every release since 2013 was rolled by Bryce Harrington, who at the time was employed with Samsung Open Source Group. Samsung shut the group down in... October 2018 [1].

[1] https://www.phoronix.com/scan.php?page=news_item&px=Samsung-...

I recall discussing in school the Federalist Papers, and in general that the authors of those papers had opponents. I believe in particular some essays by Brutus arguing for a bill of rights were emphasized. What wasn't really mentioned in my school, at least, is a specific body of work called "The Anti-Federalist Papers". That might be because the collection and name is of fairly recent origin, dating only to the 1960s. The essays themselves are of course much older, but collecting them into a numbered series of essays mirroring the way the Federalist Papers are numbered is more recent.

Matlab is still big in engineering as well. Matlab+Simulink in particular seems to have a fairly entrenched niche.

Some have at least moved incrementally in that direction by referendum. Maine, Missouri, and Oklahoma voters opted in by ballot initiative to ACA's Medicaid expansion, after their respective state governments had decided not to expand Medicaid. That's admittedly easier than designing an entirely state-level system.

Dijkstra's paper "Recursive Programming" (1960) is where he lays out his proposal for implementing recursive Algol-60 subroutines with a stack: https://www.ics.uci.edu/~jajones/INF102-S18/readings/07_dijk...

Exactly what's considered "outdoor" in that respect depends on the jurisdiction. For example DC's current regulations allow tents to count as outdoor dining as long as they have no more than one side closed [1]. Other places allow more enclosed tents.

[1] https://coronavirus.dc.gov/page/coronavirus-new-and-expanded...

There are also TSP-specific heuristics that work well in practice on a lot of large instances, often finding the optimal solution pretty quickly (but with no guarantees of optimality). The first I believe was the Lin-Kernighan heuristic from 1973 (the same Kernighan as the 'K' in K&R C, incidentally). There are fast implementations of some modern improved versions, e.g.: http://akira.ruc.dk/~keld/research/LKH/

The paper is really nicely written: https://arxiv.org/abs/2007.01409

It's lengthy and full of some quite technical sections, but those are interspersed with sections sketching the high-level intuition and proving more informal versions of the technical theorems in simplified cases. An example is on p. 19, section 3.1, "Ideas underlying proof of Theorem 3.1".

> the life of an academic is dominated by competition for grant funding

This depends heavily on where you are. I personally don't spend an onerous amount of time applying for grants. If you're at what are called "R1" universities in the US (places like MIT, Stanford, UT-Austin, etc.), then yes, your job as professor (at least in STEM) is basically to run a research lab of maybe 4-10 people, which means a lot of personnel management, finding funding to pay them, and PR (although there is some research, mentorship, etc too).

At smaller places with either a teaching orientation or more of a mixed research/teaching orientation, things work a lot differently. I personally feel quite free in my choice of research problems and how I spend my time outside of teaching. We don't have a PhD program and I don't run a "lab", just do my own research, sometimes with BS/MS students and sometimes with colleagues at other universities, so I don't have a lot of funding needs. The university does appreciate if I apply for grants now and then, but maybe 0.5-1 applications a year (I got one small NSF one, which they were super happy about). I do teach more classes than at an R1: a prof there would typically teach one a semester, while I teach two. Overall I'd estimate I spend about 15-20 hours/wk on teaching my two classes (including prep, grading, etc.), 5-10 on administration/service, and thus have about 10-20 working hours left to do whatever kind of research I'd like to do. In the summer of course the research time goes up as I don't teach.

At the moment, partly due to industry hiring away so many CS academics, I believe it's actually a quite good job to be a CS academic at a non-R1 place if you like the teaching part and the salary is sufficient for your needs/preferences. The university knows that you can leave and hiring people in this area is difficult, so if you teach reasonably well, publish anything at all, and apply for a grant now and then, they're basically just happy you're staying. This is CS-specific though; academic jobs are much more scarce in other fields, so it's less of an "employee's market" there.

How does this explain why, compared to middle-income Germans, middle-income Marylanders have low life satisfaction, feel economically insecure, and die younger?

Maybe I'm being trolled, but this is a Markov chain non sequitur response I have no way to interpret.

I'm not sure I'd say the US has consciously made that tradeoff, though maybe you aren't claiming that. If anything it's likely a side-effect, imo.

The median at any snapshot in time is materially better off, but also much more stressed out than people in other wealthy countries about whether they might end up in a very bad economic situation, due to how there is barely any bottom. Subjectively, middle-income Americans don't feel well off and don't feel secure in even what they have. For example, numbers vary based on the specific survey, but about 60-70% of Americans consistently feel high levels of stress about money. I'm not sure that's a course that would've been taken by a 50th-percentile person trying to optimize their own well-being?

One reason Scandinavian countries tend to come near the top of those "happiest countries in the world" surveys (though I don't like the term "happy" for it) isn't from any particular joyfulness, but because the typical household subjectively has very low levels of economic fear or stress. Some of the survey questions that go into that score include things like, do you feel economically secure, do you worry about losing your housing, etc. A huge percentage of even middle-class Americans spend a lot of their time worrying about that (including my family), while the majority of middle-income people from say Denmark don't consider it a big risk to worry about.

That's true! This thread I thought was about whether tuition at places like UCs has gone up because of cost increases or state funding decreases though. As far as I can tell from the budgets, it's primarily due to per-student state funding decreases, which match almost dollar-for-dollar with tuition hikes. I.e. the UC is hiking tuition to make up for declining per-student state support, not because the UC is spending more per-student to deliver the education (in real terms).

The community college system has been funded more generously, and not coincidentally, hasn't seen the same tuition hikes. You can still get an associate's degree that's almost 100% taxpayer-funded in CA, just no longer a bachelor's. (Cal State is kind of halfway in between, still much more state-funded than the UCs, but not anywhere near the per-student levels of 1990, which is why in-state tuition there is creeping towards $10k.)

In some state university systems the inflation-adjusted total funding has actually decreased.

For example, the University of California system was budgeted about $2.1 billion in state money in 1990, which is $4.3 billion in 2020 dollars. At the time, there were 9 campuses, serving 125,000 undergraduates, in a state with population 30 million. Fast-forward to 2020, and the system is budgeted $3.9 billion in state funds, which have to cover 10 campuses, serving 225,000 undergraduates, in a state with population 39.5 million.

If state funding had kept pace solely with inflation and state population, not taking into account any above-inflation cost increases, or the increase in percentage of high-school students who go to college, the budget in 2020 ought to be $5.6 billion instead of $3.9 billion, about 45% higher. If it had kept pace with both the number of students and inflation (but still no cost increases), it'd be $7.7 billion, or almost double.

My read of why it's been cut isn't that CA politicians really think the UC isn't worth funding in 2020 like it was in 1990, but more that it's been a victim of convenience in fiscal crises. CA has had a few state budget crunches over the past 30 years, and the annual UC budget is "discretionary spending" and not locked in, so is procedurally easy to cut. It was slashed significantly during both the dot-com crash and the 2008-2012 budget crisis, leaving us with today's lower level of support.