"I find it useful to think “what can I expect from speaking with the dataset of combined writing of people”, rather than treating a basic LLM as a mind."
I've been doing this as well, mentally I think of LLMs as the librarians of the internet.
One of the weirdest things Einstein discovered is that time is relative, but cause and effect are absolute.
For example, muons should decay before they hit the ground, but they don't due to time dilation. We see the time dilation when observing the muons, but the muons don't, so you would think that for us, the muons make it to the ground but for the muon it would decay too fast. However, the muons experience length contraction, so they do make it to the ground from their viewpoint as well.
So cause and effect is preserved, even though we would disagree with the muon on the relativistic reason why it is preserved.
My favorite example is if I search Google for "tide me over vs tie me over" it comes up with "Tie me over is correct". Not only is this wrong, but if you click the link, the source itself says it is wrong! The source is literally on the importance of fact checking, and Google is pulling a quote that the article uses as an example of an incorrect fact.
Google:
WARNING: It is a common misconception that the phrase “tie me over” is actually pronounced “tide me over.” Some even go so far as to say the “tide” refers to the ebb and flow of hunger, but this is not the case. Rest assured “tie me over” is correct.
Honestly I think the root problem is that universities have a degree in computer science, whereas what most people want is to learn to build computer software.
The two overlap most of the time in subtle ways where the science gives an important foundation, such as learning Big O notation and low level memory concepts where exposure helps. I've personally seen this with a smart coworker who didn't go through university and is great at programming but I'll catch him on certain topics such as when he didn't know what sets and maps were and when he tries to sleep a second instead of properly wait on an event.
However, the differences between computer science and building software are problematic. Watching my wife go through university, she's had to struggle with insanely hard tasks that will not help her at all with software, such as learning Assembly and building circuits. The latest example is the class where she's learning functional programming is not actually teaching it to her. Instead, they combined it with how to build a programming language, and so instead of giving her toy problems to teach the language she is having to take complex code she doesn't understand well that generates an entirely different programming language and do things like change the associativity of the generated language. In the end, she feels like she's learned nothing in that class, despite it being her first experience with functional programming.
On the flip side are the things that are necessary for software that aren't taught in university, like QA. For me personally, back when I was in university a decade ago I never learned about version control and thought it was just for back up. Similarly, I never learned databases or web, as the required classes were instead focused on low level concepts as Assembly and hardware. My wife is at least learning these things, but even then they often seem taught badly. For example, when they tried to teach her QA, instead of hardcoded unit tests, they made her give random inputs and check to make sure the output was correct. Of course, checking the output can only be done by rewriting all of your code in the testing files, and if there's a bug in your code it'll just get copied, so that kind of defeats the purpose. Even when the assignments are relevant there is often no teaching on them. For example, her first ever web code was a project where they told her to hook up 6 different technologies that they had not gone over in class, with only the line "I hope you've learned some of these technologies already".
Perhaps you should consider wearing an N95 while visiting the hospital. You can a pack on Amazon for $15 (even in stylish black), and they really aren't that bad wearing. In my opinion, the practice of wearing N95s in medical settings is the one response to Covid that should be permanent. Unfortunately even in peak Covid medical advisors were too scared to advise proper masking and settled for cloth and surgical masks, while KN95/N95 offers far more protection for yourself. Looking back, we could have done away with social distancing, quarantining, shutdowns, and all the other extreme procedures if we had just ramped up N95 production and told everyone to wear them in public while we were waiting on the vaccines.
" I don't think there's a JRPG more accessible than this one"
I'd argue there's one more, though I may be biased as it's what got me into RPGs - Final Fantasy Mystic Quest. It's a relatively unknown title as it was an attempt by the Final Fantasy franchise to get beginners into RPGs. It's like Super Mario RPG in a lot of ways, including seeing monsters on the overworld. However, it's much more like table-top RPGs, without mechanics such as timed hits and with more standard RPG monsters and spells. Personally I love the art style they went with in the game, it's the peak of pixel art IMO.
I agree that "throw everything in a hashmap" should be straightforward and is a good interview test. However, his further steps to "optimize" it by saying "Poor candidates load the contents of both files into memory." are terrible. Yes, that might optimize resources, but first it hardcodes the requirement that there are exactly two days breaking the solution if the requirement changes, and second it adds a bunch of finicky fragile code about "if there are two pages or more from day one or if the first page from day one is different from the page from day two".
Great candidates treat software like a business with changing requirements and code that is read by multiple people, poor candidates treat software like a math challenge where the only goal is to use as few resources as possible.
I agree. I hate how these interviews always focus so much on algorithmic time at the expense of flexibility of the code. I agree with the first part about avoiding the n^2 algorithm by using hash maps. However, making your algorithm use half the memory but hardcoding the requirement that a user visited on two days is bad design, especially when it's only halving the memory used. Also not only does it hardcode the requirements, it also makes it much more complex logic wise as you need that "if pages from day 1 >= 2 or first page from day 1 != page from day 2".
My design was to create two hash maps, one for customer to a list of days and one for customer to list of pages, though after reading the article I realized my lists should really be sets. Then you can easily account for any change to the definition of a loyal customer, as all you need to do is use two O(1) lookups and then check the size of the lists. Easy, flexible, and little room for error.
The SMTM theory is that obesity is not directly about counting calories, but more like a contagious disease due to a contaminant that causes people to be hungrier and eat more calories. Many people will dismiss this outright, but consider these things:
1. 40% of US adults are obese, which is insanely high for a willpower issue (gambling addiction is 1-2% for example)
2. The vast majority of weight loss attempts fail miserably long term, with success rates somewhere between 5-20%
3. There is precedent for this type of idea with stomach ulcers. We thought they were a psychological cause but the main cause turned out to be H Pylori bacteria
The problem is that even if they are right, it is very difficult to detect a difference between directly eating less calories and not eating a contaminant that makes you hungry so you indirectly eat less calories.
Because the story was misrepresented and it wasn't truly voted on in the normal sense.
In the US Congress, how many people are present matters to the vote. So if a bunch of Congressmen take off, it would give enormous power to the few people who remain. Because of this, there's a gentlemen's agreement that on those sorts of days, the people who are there just clock in and don't actually vote. My understanding is that the daylight savings "vote" was this type of situation where someone violated the gentlemen's agreement and voted on it without a planned vote. It doesn't really affect anything though because it would need to pass both sides of Congress and this was just one of them.
As far as why anyone be against it, I'm not sure, but it wasn't "almost" passed as the media made it sound like.
The delayed choice quantum eraser is misunderstood (even by PBS Spacetime) to invoke time travel/retrocausality. The truth is that the pattern seen on the screen is the same whichever set of detectors you send it to, and as such, there is no evidence of time travel.
The more plausible explanation behind the delayed choice quantum eraser involves Heisenberg's Uncertainty Principle, which states that you cannot obtain position and momentum to a high degree of accuracy at the same time. Einstein's EPR paradox "proved" quantum physics incorrect by setting the Uncertainty Principle against Entanglement. He did this by a thought experiment that entangled two particles, and measuring the position of one and the momentum of the other to thwart the Uncertainty Principle.
The delayed quantum choice eraser is a realization of Einstein's thought experiment. However, what Einstein didn't realize was that while the standard double slit experiment produces a pattern at the screen that gives you the momentum information, entangling two particles creates a different pattern at the screen that causes the particles by themselves to give neither momentum nor position information. The delayed quantum choice eraser initially recovers the position information by combining information from both particles, but by choosing the other set of detectors you can "give up" that information to gain the momentum information. Either way, you are still unable to obtain both pieces of information at the same time.
Source: I am writing a book on Quantum Physics and have spent months doing research and finding out that even many quantum physics Youtubers routinely say misinformation. By far the worst is the incorrect idea that a which-way detector on the double slit experiment will produce two bands as if the particles are marble-like. It doesn't, it destroys the interference pattern, producing a single slit pattern. This sounds like a minor detail, but it contains the realization that quantum physics particles ALWAYS act like waves, even after a measurement, and wave-particle duality is a misnomer.
Looking around, it does look like she's explained things better in other videos than the one I saw, so maybe I was too harsh.
The video I saw at the time is "Why is quantum mechanics weird? The bomb experiment" where she describes entanglement as a photo being torn in half and then sent apart, and then vaguely says that it is a "stronger" correlation than that.
On looking for that video though, I watched "What did Einstein mean by “Spooky Action at a Distance"?", where she describes things a bit more thoroughly with socks, where she explains that the correlations were created locally but when you measure the one particle on one side the other particle's wave function changes. She then goes on to actually explain that the sock explanation does not work as it does not sufficiently create strong correlations, which I think is a much better explanation
Looking back, I think I've learned a lot more about quantum physics since I last watched her videos. Back then, I think I saw her repeated use of explaining entanglement as a "stronger correlation" to be a cop out. Now that I've dove into the actual Bell Inequality experiment and understand how it works, I actually quite like that description as a placeholder for a very strange phenomenon that none of us really understand yet. Still, I do think she should have clarified just a bit better on the first video I mentioned, but I guess when you make lots of videos on one of the most complicated topics of all time it's rather easy to omit a crucial detail here or there.
I have substantial layman's expertise in quantum physics and while she correctly points out many mistakes that are even in textbooks, she also follows the very niche Superdeterimistic viewpoint and then presents hypotheses that follow from that viewpoint (Hidden Variables and Einstein's gloves) as if they are facts. For a science educator, this is a massive problem, and so she's become my least popular quantum physics educator even though she explains many quantum physics concepts very well.
Unfortunately, her physics videos are worse than people realize too, at least the quantum physics ones.
I've been learning a lot of quantum physics and she's a mixed bag. On the one hand, most people's (and even textbooks') understanding of quantum observation and the quantum eraser experiment are incorrect, and Sabine is one of the few voices on Youtube that calls them out instead of joins them, which is why I suspect why she is so popular.
On the other hand, she embraces the very niche superdeterminism hypothesis. This hypothesis basically says that cause and effect and reversed for any Bell's Inequality related experiment in that using uncorrelated data as inputs into the experiment causes them to become correlated because you used them for the experiment. In other words, she believes that the universe is puppeteering Bell's Inequality experiments so that we can't trust results from those experiments. This has serious issues related to non-falsifiability, where we could technically do this with any experiment we don't like the results of.
However, that's not even the worst part of it. The worst part is that she does not explain her niche viewpoints in her videos. If any of the more popular hypotheses about entanglement are correct, then Einstein's hidden variables hypothesis where entanglement acts like a pair of gloves will be disproven. However, in her videos about entanglement, she constantly refers to entanglement acting like a pair of gloves like it is a fact, because her niche viewpoint is the single viewpoint where it is not disproven.
The CDC was recommending against KN95 and N95 masks even after Omicron, so it was more like 2 years, not one month. I was also someone who wore proper masks against the advice of experts.
I agree that going with the experts is the best course of action for the general public. We need to be able to disagree with experts, but we need to be able to do it in an informed manner, with proper research to back up the ideas. However, most of the general public is unable to do so.
I was as shocked as you, and was absolutely infuriated over the bank telling me that they couldn't manually override whatever was going on. I can assure you it was a real thing that happened, and I did cancel my credit card and get a new number, if I remember I tried that at least twice.
I found the email from NPM when they fixed it, though in the email they still claim that my card details were stolen and it should be closed, ignoring that I had done that multiple times already. The email is below. Apparently there were 28 charges, so it must have been around 2 years that this was ongoing, I was dealing with some major issues at that time so I had to put it on the backburner for that time.
As far as digital wallets and virtual cards, I have none of those things. I may be a programmer, but I haven't gone techy with my finances, I just have a checking account and a credit card, and this charge kept appearing on my credit card across at least two card cancellations. Having said that, I have no idea what would happen if a fraudulent digital wallet or virtual card was set up that I was unaware of. The issue did start in 2015 though, so I'm not sure if those even existed back then.
Email from <Redacted>@npmjs.com:
"We've completed the investigation into the charges we believe linked to your card ending in [Redacted]. We've refunded each individual charge for a total of $196 (28 refunds at $7/each). You should see those credited back to your account within a few business days.
We've canceled the subscription the charges were linked to, and removed the billing details. That said, we'd still encourage you to notify your bank that the card information was stolen and that the card should be closed.
Thanks for your patience while we worked through this on our end. I understand it wasn't ideal and even frustrating at times. I'm sorry for that.
Please let us know if there is anything else we can do for you. We’ll be here to help."
Fun fact. Several years ago I started getting charges from NPM, which although I am a software developer I have never used. I cancelled my credit card multiple times, but they kept appearing each month.
I went to my bank, Bank of America, and they claimed that there was nothing they could do because NPM was using some sort of option they had to follow me when I got new credit cards. I don't know what kind of option that is, as every time I get a new credit card I have to update it with literally every other company. I also don't know how a bank wouldn't have some sort of manual override. Nevertheless, I called NPM, who said I had to talk with my bank. Eventually, after months of dealing with this loop, I threatened to leave my bank, and my bank advised me to call them and threaten to get the BBB involved if they didn't fix it, and a few days later NPM admitted it was an error on their end and reversed all of the charges.
To this day I wonder what kind of shady thing NPM was doing to not just charge someone who had never been a customer of theirs, but to follow them across cancelled credit cards.
There are, and it turns out that this is a significant mathematical concept.
The integers between 0 and infinity are defined as "countably infinite". Other infinities are considered countably infinite, or the "same" infinity, if and only if you can arrange it in a list such that each item in the list pairs to an integer in our 0 to infinity list. So the set of even numbers is countably infinite because for every i that is an even number, it pairs with the number i/2.
The decimal (real) numbers between 0 and 1 are not countably infinite, and we know this from a concept called Cantor diagonalization. What Cantor did was a proof by contradiction: assume that the numbers are countably infinite, then you can arrange them in a list. However, he then builds a number by altering the first decimal place of the first number, the second decimal place of the second number, and so on. Finally, he shows that this built number is both a real number and is not on the list. Therefore, the real numbers between 0 and 1 cannot be ordered into a list, therefore they are not countably infinite, and there are more decimal numbers between 0 and 1 than integers between 0 and infinity.
"at maximum, the spectre of AI contribution might virally taint the entire film" - which would also mean that anyone who used GitHub Copilot would not be able to copyright their software code.
I don't see programming going away for this reason. Think about it, if you have to carefully describe what you want to do to an AI - you are just writing a program. Only a program is deterministic and will do what you tell it to, whereas an AI may or may not.
The future that I see, coding and AI are divided into two camps. The one is what we would call "script kiddies" today - people who don't understand how to write software, but know enough to ask the right questions and bodge what they get together into something that mostly works. The other camp would be programmers who are similar to programmers today, but use AI to write boilerplate for them, as well as replace Stack Overflow.
I've been doing this as well, mentally I think of LLMs as the librarians of the internet.