zodiac

My guess: the definition of the BB function relies on the fact that a TM either halts or does not halt

Why is there no incentive to create artificial scarcity in the "water-delivery-and-treatment" market?

It sounds like you have a problem against privately owned water rights (the exclusive right to extract water from some region and then sell it), not against water futures per se?

FWIW I'm sympathetic to that position (in some form), it would just be better to be more precise... people are acting like "water is a basic human right" is a principle that can straightforwardly be used to derive the rest of our water policy.

I guess I don't understand the difference between paying for "water-delivery-and-treatment-infrastructure-and-operational-costs", and paying for "water". I can't fufil these water futures by just transferring legal ownership of a piece of land (that I owned) that contains a saltwater lake.

Even after defining what the difference is... I'd be interested in understanding what makes trading in "water-delivery-and-treatment-infrastructure-and-operational-costs" OK, but trading in "water" not OK

I'm not sure what you're saying - is it that "banning water futures" will likely accelerate it, or that "water futures" will accelerate it?

> trading with water is... dangerous...

Do you pay a water bill where you live? If so isn't that already "trading with water"?

> especially if we ever get to a point where some people can 'afford' water while others can't.

If we do get to a point where this is true, it will primarily be because of decreasing amounts of water sources, increasing population, etc. Banning water futures won't magically solve these problems.

Yes, I noticed this from the statement "apply that binary operation to one number from the list". They're basically treating f as a function from R to R.

The fact that I'm an ios developer

> Option types are way better and always do what you expect.

No, I can implement the exact same weird functions with option types, with signatures

sum: Array<Int> -> Optional<Int>

count_elements: Array<Int> -> Int

FWIW, I personally prefer both functions to not return Optionals, but the point is that the mere presence of Option types in your type system (and the absence of null pointers) doesn't magically eliminate bad library design

Well, I think that's analogous to the max function having signature (say) Array<Int> -> Optional<Int>

How do I sign up to be an auditing firm?

> Certainly the idea that a score somehow has more information in it than a midi recording is ridiculous

A midi file is designed to contain different kinds of information than a score, so it can absolutely contain less information (of a certain kind) than a score.

Imagine a group of notes that are attacked very close together in time in a midi file - is it meant to be a chord played without arpeggiation? A chord arpeggiated as fast as possible? A chord arpeggiated to a specific tempo?

When a composer writes (say) a chord with an arpeggiation sign next to it, it specifies that the notes are to be played in a certain order, and quickly. There are multiple possible performances that comply with this "specification". MIDI cannot encode this information (the information being the set of acceptable performances).

Perhaps this? https://en.m.wikipedia.org/wiki/Painlev%C3%A9_conjecture

> If someone has a better explanation of why subtracting relatively prime numbers repeatedly will always result in one, please leave a comment or reach out and help me understand better.

You're looking for a proof of Bezout's identity, or something in the spirit of that theorem :) to be honest, I haven't seen a proof of it that I "really really like", although you might find one

I prefer to use a dot like you, but find calculations like "let x=5, then 3(1+x) = 3(1+5) = 3(6) = 18" acceptable, because each equation is just a lock substitution

If you define GCD this way then writing gcd(a, b) = c is an abuse of notation :)

I'm a Singapore citizen and I've used my Singapore passport to get the tourist SIM cards from M1 before, did some store refuse to sell it to you?

I hear you about those language pairs that have to be round-tripped through a third language. I completely agree, too, that the big open questions in NLP are all about understanding meaning, semantic content, pragmatics, etc rather than just syntax.

I don't think that "NMT and similar techniques" ignore meaning by design though. What they do do by design, compared to expert systems etc, is avoid having explicitly encoded knowledge (of the kind SHRDLU had). Take word2vec for instance, it's not NMT but fits into the "statistical NLP" description - its purpose is to find encodings of words that carry some semantic content. Now, of course it's very little semantic content compared to what an expert could plausibly encode, but it is some semantic content, and this content improves the (subjective human) evaluation of NMT systems that do use word2vec or something similar.

Also, we should carefully distinguish "prior knowledge" as in "prior common-sense knowledge" and "prior linguistic knowledge". The end-to-end trend eschews "prior linguistic knowledge", while current NLP systems tend to lack "common-sense" knowledge, for rather different reasons.

End-to-end training tends to eschew prior linguistic knowledge because it improves (subjectively evaluated) performance in real-world tasks - I believe this is true for MT as well, but an easier example if you want to look into it is in audio transcription. I don't think there's a consensus about why this happens, but I think it is something like - the previously way people were encoding linguistic knowledge was too fragile / simplified (think about how complicated traditional linguistic grammars are), and if that information can somehow be learned in the end-to-end process, that performs better.

Lacking "common-sense" knowledge - that's more in the realm of AGI, so there's a valid debate about to what extent neural networks can learn such knowledge, but the other side of that debate is that expressing common-sense knowledge in today's formal systems is really hard and expensive, and AIUI this is also something that attempts to generalize SHRDLU run into. But it is definitely incorrect to say that it's ignored by anyone by design...

BTW, the biggest improvements (as subjectively evaluated by me) I've seen in MT on "dissimilar languages" have come from black box neural nets and throwing massive amounts of (monolingual or bilingual) data at it, rather than anything from formal systems. I use deepl.com for Japanese-English translation of some technical CS material, and that language pair used to be really horrible in the pre-deep-learning days (and it's still not that good on google translate for some reason).

> Look at the promise of AI: "machines that think like humans!". Look at the reality: "We're generating value!". OK.

OK, I hadn't realised we had such different implicit views on what the "goal" of AI / AI research was. Of course, I agree that the goal of "having machines think like humans" is a valid goal, and "generates value by serving production traffic" is not a good subgoal for that. However, this is not the only goal of AI research, nor is it clear to me that for e.g. public funding bodies see it as the only goal.

I use MT (at least) every week for my job and for my hobbies, mostly translating stuff I want to read in another language. I love learning languages but I could not learn all the languages I need to a high enough level to read the stuff I want to read. The old non-NMT approaches produced translations that were often useless, whereas the NMT-based translations I use now (mostly deepl.com), while not perfect, are often quite good, and definitely enough for my needs. Without NMT, realistically speaking, there is no alternative for me (i.e., I can't afford to pay a human translator, and I can't afford to wait until I'd learned the language well enough). So how can you say that AI "remains completely useless"?

Basically, you have implicitly assumed that "make machines that think like humans" is the only valid goal of AI research. And, from that point of view, it is understandable that evaluating NMT systems for how well they approach that goal using human evaluations, has many downsides. However, while some people working on NMT do have that goal, many of them also have the goal of "help people (like zodiac) translate stuff", and in the context of that goal, human evaluation is a much better benchmark target.

The subjective human evaluations used in research are blinded - the humans rate the accuracy of translation without knowing what produced the translation (whether a NMT system, a non-ML MT system, or a human translator), whereas the football fans in your scenario are most definitely not blinded. There are some criticisms you could make about human evaluation, but as far as how well they correspond to the real-world task, I think they're pretty much the best we can do. I'm very curious to know if you actually think they're a bad target to optimize for.

More to the point, you still have yet to show that NMT "serves no other purpose than to show how well modern techniques can model large datasets", given that they do well on human evaluations and they're actually producing value by serving actual production traffic (you know, things humans actually want to translate) in Google Translate. If serving production traffic like this is not "serving a purpose", what is?

> How do you mean that machine translation outperforms non deep learning based approaches at the real world task? How is this evaluated?

There were two ways that performance can be evaluated that I had in mind:

1. Commercial success - what approach does popular sites like Google Translate use? 2. Human evaluation - in a research setting, ask humans to score translations - this is mentioned in your link

Yeah, I know about how current automatic MT benchmarks don't reflect "user satisfaction" very accurately, and that it's an open problem to get one that is serviceable. However you make it sound like all deep learning solutions to tasks perform well on the task benchmark but poorly on the real-world task the benchmarks try to approximate, whereas that's not true for MT - they are bad at the benchmarks, but outperform non-deep-learning-based translation approaches at the real-world-task.

On the subject of benchmarks, how about speech transcription? I was under the impression that those benchmarks are pretty reliable indicators of "real-world accuracy" (or about as reliable as benchmarks are in general)

For me, codeforces is fun the same way chess is fun. Yeah, the mechanics look dry on the surface, but lots of variation under the surface, and lots of chances to link concepts by having an aha moment during the contest or afterwards in analysis.

> It's no stretch to imagine that an entity of the size of Google or Facebook could achieve something considerably more useful, still in a limited domain. But this has never even been attempted.

How do you know if no one has attempted it, or if all attempts so far have failed to achieve their goals? One of the claimed downsides of "hand-crafted systems in limited domains" (i.e., something like the SHRDLU approach) is that they would take too much effort to create when the domain is expanded to something even slightly bigger than SHRDLU's domain, so a lack of successful systems could be evidence of no one trying, or it could be evidence that the claimed downside is indeed true.

The fact that a working system for the limited domains of e.g. customer support or medical diagnosis would be worth a lot of money suggests to me that they must have been tried, but that nothing useful could be built, and we didn't hear about the failed attempts, meaning that those domains (at least) are too big for hand-crafted systems to work.

> Yes, it is faffing about. Basically NLP gave up on figuring out how language works and switched to a massive attempt to model large datasets evaluated by contrived benchmarks that serve no other purpose than to show how well modern techniques can model large datasets.

It is inaccurate to say that all benchmarks are useless. For instance, language translation (as in Google Translate) is a benchmark NLP task, but it is also something I personally use at least every week, and deep learning based solutions beat handcrafted systems by a lot for this particular task (speaking as an end user who has used systems based on both approaches). The same comments apply to audio transcription (e.g. generating subtitles on YouTube) as well.

Your second sentence is true, but the first one doesn't make sense. A car travels at about 0.000001% of the speed of light, yet when we measure car speeds, they depend on the frame of reference.

Yeah, I would say that understanding conjectures often requires knowing some "background lemmas" in addition to knowing the definitions of the objects used in the definition, for this very reason. As another very simple example it can be shown that you can replace ℕ by the set of integers greater than 3, and the resulting conjecture is equivalent.

Really? I own a Huawei and AFAICT, the OS / services (eg the kernel, android, Google play services) are basically the same as what would run on eg a Google pixel.

I would guess that buying an app with the intention of harming the app creator by doing a refund is against the ToS, and I think that's fair (since the purpose of the refund mechanism is to allow the buyer to change their mind for some "good faith" reason - even something like a change of mood is fine, but "to harm the creator" is obviously not)

What does this have to do with the bulk of the article (which is about algorithms the author used at his jobs, and only tangentially (IMO) about the use of algorithmic questions in interviews)?