It should be at least "Put country first, then ask for zip code".
08036 Barcelona, Barcelona, Spain
08036 Hainesport, New Jersey, USA
It does make some sense to go from most general (country) to more specific (address) instead of the opposite, so the form can implement better auto completions.
However in practice postal addresses are not that clean so simple dumb forms are usually better, because it's easy for exceptions to be missed by auto completion algorithms. I would be very pissed if I had to type the zip code above and then correct the wrong guess on every website.
Either you are helpful and guess correctly or you don't guess at all.
I find that the tricky part of a good data analysis is knowing the biases in your data, often due to the data collection process, which is not contained in the data itself.
I have seen plenty of overoptimistic results due to improper building of training, validation and test sets, or using bad metrics to evaluate trained models.
It is not clear to me that this project is going to help to overcome those challenges and I am a bit concerned that if this project or similar ones become popular then these problems may become more prevalent.
Another concern is that usually the "customer" asking the question wants a specific result (something significant, some correlation...). If through an LLM connected to this tool my customer finds something that it is wrong but aligned with what he/she wants, as a data scientist/statistician I will have the challenge to make the customer understand that the LLM gave a wrong answer, more work for me.
Maybe with some well-behaved datasets and with proper context this project becomes very useful, we will see :-)
I have been using Home Assistant for more than 5 years. The stability of the system has improved a lot in the last year. I don't recall the last time I had to reinstall or restore a backup.
At the beginning (0.7 or maybe even earlier) I remember to have to reconfigure or reset my instance a few times a year. Those times are long gone.
Another physicist here, not an expert. This is my understanding of the work:
The paper (and related papers) aims to unify classical electromagnetic (Maxwell equations) and gravitational forces (Newton laws) in a new proposed framework, the informatons.
Assuming the maths check out, my understanding is that the author proposes the informatons, and successfully rewrites Newton and Maxwell's equations in terms of the informatons.
If I understand correctly, the author believes informatons are a better theory because electromagnetic and gravitational fields are mathematical constructs, just like informatons, but with the informatons you have one framework capable of explaining both phenomena. It sounds to me like a unification theory for both gravitational and electromagnetic fields.
Having one theory for all fundamental forces has been a goal of theoretical physics for a long time.
There are four forces to be unified (strong nuclear, weak nuclear, electromagnetic and gravitational). The weak and electromagnetic forces have already been unified (awarded a Nobel Prize in Physics, 1979). The electroweak unification has been experimentally proven. https://en.m.wikipedia.org/wiki/Electroweak_interaction
It seems to me the author aims to unify gravity and electromagnetism. The proposed theory may be mathematically correct but to my understanding it has two big limitations:
(1) there is no experimental setup devised to verify the existence of these informatons
(2) classical electromagnetism has already been unified with the weak nuclear force.
There have been several attempts to unify classical forces (electromagnetism and gravitational), but since the electroweak unification in 1979 physicists do not tend to focus much on a electromagnetic-gravitational unification. And physicists have a reason: If you already have the electroweak unification, why would you want to find a theory that allows you to merge electromagnetic and gravitational forces without considering the weak nuclear force?
It is hard for me to find the author contributions valuable without:
(a) A focus on how informatons should be experimentally validated, just like it happened with the electroweak unification. This is a major problem for this theory, in my opinion.
(b) How does the author deal with previous proven unification work, for instance the electroweak unification?
(c) In general, the author does not seem to be able to put his work in context of other unification theories. This is usually a red flag to the community because it implies that, as a reviewer, I can't assume the author knows the state of the art of the field.
Having said this, I am no theoretical physicist and I am no expert, so I will be very happy to be corrected.
> s. Upon analyzing the distribution of
Docker Desktop for Windows and Docker Desktop for macOS, it was discovered that running the
Docker environment on these requires a lightweight virtual machine that emulates the Linux system.
"it was discovered" as if the VM behind the docker engine running on those operating systems was hidden to the human race.
> The dynamite plots.
Please don't create dynamite plots. Use a box plot, or plot the dots of the different repetitions with some jitter to avoid overlap.
> The evaluation
As reported in other comments apparently OSX cheats when saying it's writing to disk. If the authors wanted to compare docker executions it would be useful for the discussion to see if the difference in performance was due to the docker overhead or to the operating system, by showing what's the write performance of a non-containerized program across operating systems.
In my opinion the article should not have been published as is. It should have been rejected or asked for major revisions. MPDI Applied Sciences is quite a bad journal already for "Computer Science Applications" https://www.scimagojr.com/journalsearch.php?q=21100829268&ti... so it's not surprising to see this published.
Did the authors use an LLM to write or improve the text? I have no problem with that but I feel I'd like to know how much work is LLM based before reading.
Using intervals for measurements has some limitations. But for many use cases we do not need more than intervals, so it's nice to have convenient tools for them. Intervals are a convenient model.
That's because measurements are complicated.
You use a ruler (or some other instrument) to measure something and get a value x.
You are happy.
Then for some reason you decide to repeat the measurement and you get a slightly different value. And problems start.
You decide to write down all the values you get. You are happy again.
Shortly after, you realise you have to use those values in calculations and you just want "one representative value", so you take the average or "the most common value" or some other aggregation, use your intuition!
Things start to go wrong when you have to take a decision by setting some threshold like "I do this if my value is above a threshold". Because the actual value may be different from your averaged number.
So you take the standard deviation and call it the uncertainty x±s.
But one day you realise that your measurements are not symmetric. You start by saying "instead of x±s, I will use different upper and lower bounds to define an interval".
For instance some things are measured on a log scale and you have a measure like 100±"one order of magnitude" which is "100, but may be between 10 and 1000".
Then you add a confidence, because you are not 100% certain you actually are in that range. Your measurement becomes "with 95% confidence I can say the measure is in [10,1000], with an expected value of 100".
Then you want to combine and aggregate those intervals and you realise they within the intervals their regions are not uniform, you actually have a probability distribution.
In the simple case is a Gaussian distribution, described with mean and variance. It can also be a binomial (a "p out of n cases" scenario). Or a lognormal like un our 10-1000 example.
And now for each measure you take you need to understand what probability distribution it follows and estimate its parameters.
And that parameter estimation is a measure, so it has confidence intervals as well.
At this point adding two measurements becomes not so easy anymore... But don't panic!
The nice part about all of this is that usually you don't care about precise error estimates, because you can live with bounding errors covering a worst case scenario.
And you can use the Central Limit Theorem (sometimes it is abused rather than used) to simplify calculations.
It is a rabbit hole and you need to know how deep you want to dig. Intervals are usually convenient enough.
I would believe those results if they had published the methods (including everything they were correcting for, etc) BEFORE starting the data collection.
Otherwise I can argue that once you have all the data it is feasible to test many different combinations of variable corrections, age groupings, tattoo sizes, etc. until you find one scenario (the one you publish) with "statistical significance".
I had a quick glance at the article and the authors do not discuss any multiple testing correction method. The lack of such discussion makes me think that they were unaware of such problem and they tested multiple hypothesis until they found one "statistically significant". This is called cherry picking. https://xkcd.com/882/
I can think of two alternatives to the cherry picking hypothesis that might make me believe their conclusion will hold:
1. They had the statistical analysis plan decided from the beginning up to the smallest detail. They followed it by the book and they found the published result without exploring anything else, so there were no other hypothesis to correct for multiple testing for. This is feasible, but seeing the significance and effect size it seems this would be a very risky study design, since the effect size they see is rather small. Since it's so risky, I find it unlikely.
2. The result holds regardless of forcing minor variations to all those corrections. This means there would probably exist a simpler analysis plan, without so many corrections, that presents compatible results with the published plan. This is unlikely in my opinion, because if that were the case I would expect a simpler story in the paper, or a larger effect size.
Maybe an independent group of researchers believes in these results and decides to reproduce the study to confirm it. If this happens I hope they follow the same statistical analysis plan published in this paper, and I hope they can publish their findings in the same journal, even if they can't get a "statistical significance".
The closest solution I remember to this happened roughly 15 years ago when there was a browser extension that detected pgp on Email bodies and was able to get public keys and encrypt/decrypt emails automatically or with a single click, if I recall correctly.
It felt a small team or a single person project. There was a cat and mouse game being played against Gmail interface changes, with the encryption being broken every now and then. It required me, a random user, to trust the extension developer(s) with my Gmail and gpg keys. I had fear of having an xz backdoor situation compromising my Gmail account and gpg keys.
And for this extension to succeed I also had to convince all my Gmail friends and contacts to use it, and convincing them to trust the developers of the extension as I had done.
While I felt confident that the extension was perfectly safe I didn't dare to convince my family and friends to use it because (a) I could not guarantee that the extension wouldn't be compromised in the future and (b) the chance of the extension being broken in the future due to a Gmail UI change was close to 100%. Also the more users the extension had the higher the chances for someone to attempt to infiltrate to attempt a backdoor...
If I recall correctly, eventually the developer got tired of the Gmail UI changes and increasing demands of users and moved on, stopping the development.
Anyone with time could try to develop such an extension, but there were drawbacks back in the day...
Fully agree. Switching from the time domain to the frequency domain is just like switching coordinate systems.
If you have a signal with a single narrow peak in the time domain you can represent it in a very compact or sparse way just using a delta at the position of a peak. If you try to represent it in the frequency domain you will not find such compact representation. Similarly if you have a sinusoidal signal in the time domain, you won't get a compact representation there, but you will get it in the frequency domain where you just have a couple of deltas.
Time and frequency are two ways of representing the same thing. Sometimes it's easier to represent something in one domain, sometimes it's easier in the other domain.
It can be proven that anything bounded in the time domain will be unbounded in the frequency domain and viceversa. So compact stuff in one domain always spreads when represented in the other domain.
Beautifully, quantum mechanics tell us that position and momentum are conjugated variables (like time and frequency in the example above) and therefore if something has a bounded position (we know where it is) its momentum will be unbounded (we won't know its speed), and viceversa.
That's the main idea of Heisenberg's uncertainty principle.
When I listen to speech synthesis systems from the 80s, like the one from Stephen Hawking, I feel they are easy to understand but "robotic", they sound all the same in a "robotic" sense.
LLMs seem to suffer from that as well. The length of the answers has lower variance than the length from humans. The number of paragraphs, the length of each paragraph, the structure of the paragraphs... All that is kind of predictable, so when I read a text that does not have that structure I usually don't think it's written by an LLM. However if the text has that structure then I can suspect it may come from an LLM. I believe in some months we will all be used to the LLM style and we will get much better to identify it... And then chatbots will probably be changed to sound less robotic...
When I had to study and memorize some text at school I used to remember the page where the text was written and then mentally read it. Some of those pages are still in my head (for instance multiplication tables).
Once I was on an exam and I could not understand my own writing in the page I was remembering because I had written it too small on the corner. It was frustrating to not be able to answer the question. Afterwards, when I went to my real notes and I struggled to understand what was written there. I was happy that my memory image was accurate although frustrated for not managing the space in the page properly.
I thought everyone could remember things in this way.
Sometimes I have to write things down to see them and remember them, because mental speeches are harder (and less efficient) for me to remember.
I guess there are many ways to learn and remember stuff we just have to find the one that works better for us.
Being a teacher should require knowing about all this learning diversity I guess.
Last week I was travelling on a different country. I was mostly without WiFi and I bought a prepaid SIM card with some GB. TikTok ate 2GB of "background data" in two days, even if I did not open the app.
So it's not hard for me to guess how they do it...
(Yes, background data is now disabled for TikTok on my phone... I learned my lesson)
08036 Barcelona, Barcelona, Spain
08036 Hainesport, New Jersey, USA
It does make some sense to go from most general (country) to more specific (address) instead of the opposite, so the form can implement better auto completions.
However in practice postal addresses are not that clean so simple dumb forms are usually better, because it's easy for exceptions to be missed by auto completion algorithms. I would be very pissed if I had to type the zip code above and then correct the wrong guess on every website.
Either you are helpful and guess correctly or you don't guess at all.