throwawayiionqz

The challenge of modern academia in certain fields is not to publish but to be read by others. Everyone is so busy publishing that very few papers get decent readership (retweets and citations happen mostly without reading the substance).

I would rather have an upper limit on the number of papers one can publish in a year than more avenues to publish unsubstantial findings.

Thanks. So I go for a haircut, if I pay in EUR the hairdresser forwards the VAT to the government. If I pay in the local currency for the same haircut, the hairdresser keeps the VAT. Same if I buy fruits from a local producer (how is that "coup de main"!?)

This very much looks like VAT or other tax evasion scheme. Maybe users of the currency should realize the associated risks.

If the government shuts it down for fraud, it will be difficult to convert the local currency back to euros and the claim that is pegged 1:1 to the Euro might be be correct after all.

How is VAT paid? The government is clearly not accepting the local currency.

So VAT must be paid in euro for purchases in the local currency? I am wondering how this eventually works out, especially the day when selling your local currency into euros is not anymore possible or pegged 1:1, while VAT must still be paid in euro.

National security. Why would the EU let some foreign country have access to all their citizens' private and work email, social media profiles and e-commerce purchases? Especially when it has been proved that that country's intelligence agencies have access to all of FANNG's data on EU citizens through various intelligence programs.

Can't wait the EU force FANNG to sell their EU activities to native EU firms

The next problem is the testing infrastructure. Few labs (if any) do PCR tests with results guaranteed in the next 72h in major US cities.

What if test results are late? Should you delay and take the first flight after getting the results? Now the test was taken 4 days or more before the new flight. Should You take another test?

Airlines or countries that do not want to welcome travelers without a recent, <72h PCR test should set up the necessary infrastructure in the city of the departure airport.

And the cherry-on-top: Once you are part of a research unit at big corporation, your annual bonus depends on your number of publications. Or why the pursuit of incremental but well marketed results never ends. Or why papers with 6+ authors among which only 1-2 substantially contributed is the new normal.

Broadband and mobile prices are far more reasonable in Europe than in the US. Unlimited data+calls for <20EUR has been a thing for several years now in Europe, and broadband for <30EUR as well.

Also the EU banned roaming fees and GDPR prevents the data reselling and privacy violations that carriers and ISP in other countries could introduce at some point.

On the other hand the US is unable to keep broadband companies from arbitrary price increases or per-building monopoly. Health care prices are also out-of-hand, it seems the US is happy to let their citizens getting ripped off as long as someone is making a big profit.

There is a lot of anecdotes in the comments about these interview processes.

The ones with the best data are the company themselves, especially those who interview thousands of candidates a year. These companies (e.g., all FANG) have enough data to assess which interviewer, which interview questions and which interview process have positive correlation (and even causation) with future employee performance. And they can optimize the pipeline to interview most candidates with the interviewers that can actually assess future employee performance for instance.

Either these companies have already done these optimizations and analyze the trove of data they have, and the process is what it is for a reason. Or they haven't yet and there is still much room to optimize the process, and applicants have a chance to see those processes updated for the better (although here better means better for the company, not necessarily better for the applicant).

Indeed, if you wish the sequence to be defined at 0, eru's solution is better (and appeared earlier while I was typing)

The "better" recursive solution computes fib(n) and fib(n-1) at the same time. Linear time recursion!

    def f(n):
        if n<=2: return (1, 1)
        else:
            a, b = f(n-1)
            return (b, a+b)

    def fib(n):
        return f(n)[1]

There should be much smarter moves to work with porn sites towards better sex education.

Force porn websites to do age verification on their models (most high traffic site currently do).

For all videos and especially those that display acted scenes where consent is unclear, prominently display next to the video in English+local language that the video actors and actresses are performing willingly consentual sex acts, even if the video scenario suggests otherwise for some fantasy's sake.

We only paid $6k from HSA (maxing out the out-of-pocket) but that Mt Sinai hospital charges $80k (after adjustments) and the insurance pays the rest. I should add there was a C-secyion that barely added $10k to the bill.

I know, couldn't believe it either

Private hospital in NYC: Just the hospital "room and board" line is north of $40k: $13k per night of "enhanced nursery" for the newborn and around $10k per night for Mom. That's before all the doctors add their services.

No idea how you can make it lower than $10k :) Where was that?

2 nights at the hospital for delivering a healthy baby may cost $50k-$100k to the insurance for healthy baby and mom.

The fact that hospital/physicians charge as much and insurance accepts to pay is not due to some unhealthy habits or chronic diseases in the US.

Edit: of course your point stands and the higher rate of chronic diseases makes it worse.

To add some context, the "critical" reviews of the product [1] describes multiple instances of the problem.

[1]: https://www.amazon.com/Aquaphor-Healing-Ointment-Advanced-Pr...

Edit: Searching with "yellow" or "fake" brings more reviews with the same issue. In the mean time parents keep receiving it and applying it on their newborn.

Before calling the AG on this, how do you make sure that fake is fake in the first place?

The major hurdle is that you do not even know and cannot know. Maybe the product is actually legit and the visual difference comes from temperature or whatever, and the rash appeared for a different reason. One avenue may be to contact the manufacturer with the lot number.

If you try a new product and baby develops rashes or allergies, you know not to buy this product again. With Amazon you can't even conclude anything about the official product because you might have used a knockoff.

You can just conclude to stay away from Amazon for baby care products.

I ordered a well-known baby-bottom cream. The product gave her never-expanding rashes for a few days. Then I realized the color slightly differs from the same cream bought at the local pharmacy, and that many 1 star reviews also report knockoffs version of the cream. In some cases review pictures show that the packaging itself is clearly fake.

I write a 1-star review explaining the above, and saying that the anxious parents are now in an unbearable situation after applied a fake, potentially harmful and untraceable product to their newborn bottom.

The review does not respect community guidelines and won't be published.

Some reference is Exercise 84 (Cantor function) in Tao's lecture notes at https://terrytao.wordpress.com/2010/10/16/245a-notes-5-diffe...

The results before/after that result have other niceties: a one sided fundamental theorem of calculus for monotone functions (even though the Cantor function has 0 derivative!) or the fact that monotone functions can also be decomposed as the sum of continuous monotone function and a jump function such as Cantor's function.

If subgradients are enough (-1 is correct subgradient at 0 in your example) then there are valid approaches for AD subgradient, see https://arxiv.org/abs/1809.08530

I am wondering if the NSF program officers responsible for overseeing the grants should be contacted as they would be in power to make the University and the PIs accountable and/or start their own investigations.

They can be easily found in grant webpages, eg https://www.nsf.gov/awardsearch/showAward?AWD_ID=1900713

This does not make sense.

Before scientists started reviewing each other, who at the journal was judging the work of scientists to be published? Scientists?

It's even easier to trick the authors into believing you (the reviewer) is someone else.

E.g: ask the authors to cite two papers of another researcher (as was mentioned somewhere in the thread as making it easy to guess who the reviewer is), ask to compare to results of that researcher more thoroughly, use some writing style typical of that researcher country of origin, etc.

It's especially easy if you are familiar with works from the other researchers for instance because you also reviewed their work.

It's not in-class interactions (party? Networking? Labs?) versus being alone in a room watching lecture videos.

If students don't have to live on campus to get the credits and the degree, it opens a world of possibilities. They can work part-time/intern in a real company and network while completing their online courses. Receive salary at 18, network with other interns and older employees at the company, participate in industrial grade lab experiments if that happens at the company.

Say Google would open "3y part time internships" allowing students to enroll in a credentialing University and complete courses and get a degree online while working at the company. The in-campus experience doesn't sound very compelling now.

There are two NYC studies with likely over-sampling phenomena:

- the NYC antibody study one you mention, done at shopping centers, indeed likely over-represents people going out. 20%-21% of that population has antibodies.

- the SARS-CoV-2 testing study with pregnant women [1], tested just before delivery. Among those, 13% tested positive. It is reasonable to expect that this study over-represents subjects that barely go out: pregnant women to go out as little as possible to protect themselves and their future baby.

Because the sampling over-representation is opposite in the two studies, the truth is likely in between. in terms of antibodies, it is likely that the pregnant women that tested positive weeks ago have now developed antibodies or will do so in the next 1-2 weeks. Among pregnant women, the 13% also ignores women that had already developed antibodies before delivery; another phenomenon that may push the truth above 13%.

[1]: https://www.nejm.org/doi/full/10.1056/NEJMc2009316

I am simply trying to slow the spread of non-sensical numbers. The total population divided by number of deaths provide a factual lower bound and arguing against that is not sensible.

Flaws in the Stanford study has already been highlighted extensively, see for instance https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaw... and the insightful comments. The issue is not to take it up with the authors but whether the authors might listen and take back what they claimed on popular political TV shows. [1]

[1]: https://www.youtube.com/watch?v=6NjCitwKJSQ

Number of deaths in NYC divided by total population is 15,000/8,000,000. This already puts the fatality rate at at least 0.2.

But the consequences are far from over and the number of deaths will keep climbing, so your "fatality rate may be as low as 0.10-0.20%" is completely off the mark.

edit: I am sorry if my tone was interpreted as confrontational. I otherwise agree with the parent's post that we need much more antibody testing.

The derivative of x->x (identity) should be 1 regardless of your definition of derivative.

Subgradients are only applicable when summing convex functions. Here relu(x) - relu(-x) is a sum of a convex function and a concave function.

I am wondering what is the state of things for fixing gradients are kinks, or if there is no hope for general automatic differentiation libraries.

    import jax.numpy as np
    from jax import grad, jit, vmap
    from jax import random
    
    def relu(x):
      return np.where(x>0, x, np.zeros_like(x))
    
    def identity(x):
      return relu(x) - relu(-x)
    
    derivative_identity = grad(identity)
    derivative_identity(0.0)

returns 1.0 or 0.0? It currently returns 0.0. (Edit: typo in sign)

I teach my regular class as you did: sitting at the desk, facing the students in the room, writing on a tablet whose screen is projected behind me.

1) I get a lot more eye contacts with the students this way versus using the blackboard, because I always face the students and never face the blackboard.

2) Beyond streaming the screen, there exist software that records and share what was displayed in previous slides. Students watching the video can then very easily (alt+tab) access another window where they can scroll to check previous slides. OneNote lets you do this for instance by sharing a notebook; the students can either access the notebook on a webpage (sluggish) or on their own OneNote app on a tablet. I typically have 1 notebook for each course and the notes taken during class are updated almost instantly (few seconds). Sadly OneNote has drawbacks: importing a PDF rasterizes it and the result is blurry/pixelated, and the latency for updating the notebook online is few seconds, so not good enough for online streaming.

3) Webex or similar conference software lets each participant send a video stream or a screen sharing stream. I intend to try this setup in the next few weeks: (a) me sharing a selfie video stream (b) me sharing my iPad screen where I write or annotate slides in OneNote, (c) the OneNote online notebook that students have access to in another window with all previously annotated slides, (d) hopefully the students streaming a selfie video of themselves to get visual feedback in the Webex meeting.