simiones

There are more layers to this problem, if we want to get into the details. The LLM is defined in terms of floating point operations, and those are not actually fully deterministic, on most hardware and in most performant implementations.

IEEE 754 only specifies precision requirements for certain operations, not precise bit patterns (e.g. for exponentials). So, at least in principle, the same hardware performing the same operation could produce different results at different times, as long as they are close enough to the theoretical answer. I'm not sure if any hardware actually works like this.

IEEE 754 also specifies that many of the basic arithmetic operations are not associative - so any reordering (which is common when batching multiple queries at the same time) will introduce indeterminacy from the perspective of your own query (that is the result for your query will change depending on what other query happens to be processed at the same time, which is not under your control).

Finally, even if we take the case when a query is processed alone, and even if one particular hardware is completely deterministic, the result will be different on different hardware - which can again look like non-determinism if you're sending your query to a load balancer.

So, the math for LLMs is deterministic in theory, but implemented with non-deterministic approximations & optimizations in practice, and their results are then normally used only as a probability distribution to be sampled from.

This sub-thread was about cloud environments, where different requests may be served by different hardware. And it's in fact very likely that there will be a mix of different hardware from different vendors, in any particular LLM cloud for now.

In almost all forms of math, the value n/0 is undefined. It's definitely not infinity, for two reasons - depending on the value of n, it can be negative; and neither info nor -inf are numbers, so they can't be the result of an equation (unless you look at transfinite equations).

What you can do in math is talk about the limit of a series of fractions as the denominator approaches 0, and that's where you get some relation to infinity or -infinity. But the limit can also be any other number, if the numerator also gets closer to 0; or it can not exist, if the function oscillates.

The very concept is quite culturally contingent - both in how much buyers would care about getting such a service, and in how much celebrities would be willing to engage in this type of fan service. Going global would also require a huge amount of celebrity contractors, a big problem for scaling.

There's a huge difference between purely intellectual subjects, like organic chemistry and integrals, and the mix of emotional and intellectual depth of a novel. A lot of the meaning of literary works is built on top of shared human experiences, just like the meaning of integrals is built on top of more foundational math. However, we don't have any good way (and certainly make no effort in school) to teach this shared human experience to pre-teens and teens.

A good part of the value of some of these works basically comes from recalling similar feelings you felt in situations similar to the characters, maybe comparing your actions at such times to theirs, or the reactions of other people you knew, etc. It's simply not possible to experience this part of the work as a teen. Perhaps one of the clearest examples of this limitation is in Lolita - the nature of the relationship described, the power and life experience differential, the contrast with the reader's normal interactions with children - are impossible to be conveyed to or truly empathized with by teens.

I would say that Checkhov is mostly known for his plays, not for short fun stories.

I don't think it's conceptually difficult, it's just hard (for some of us, at least) to remember the names of a culture you haven't been exposed to. I speak a language with slavonic influence, so I didn't have this problem with Russian literature, but I remember vividly how hard it was to remember the names of characters in the first anime I watched, just because I was so unaccostemed to Japanese names (even though they were clearly very distinctive).

The recall of words you aren't familiar with tends to be pretty poor. This is also visible in how hard it can be to build vocabulary when learning a new language, and how you can completely mix up words at that stage - there's nothing about names that makes this any easier.

You mean GNU Hurd, I guess.

There are many crappy design decisions in Unix, Posix, Linux, MacOS, that have been known and worked around for decades (see the Unix Hater's Handbook). One example would be async IO, which has been famously bad in Unix, and Linux alone has tried 2 different generations of solutions - with the newer one, io_uring, being suspiciously similar to Windows' decades old IOCP. Fork/exec is infamous as a needlessly complex process creation API, requiring huge effort in the kernel just to handle the most common case of simply launching a new process. The traditional Unix security paradigm of simple user based file permissions is extremely weak and has required numerous solutions on top to handle realistic security scenarios - e.g. to protect a user's files from being tampered in unexpected ways by a process launched by that user (SELinux, jails, namespaces, etc).

I agree that Windows is crappy, but that doesn't mean that Linux and MacOS aren't also crappy in their own ways (not to mention iOS, Android).

I think the exclamation point at the end of GP's message is intended to indicate sarcasm - GP also doesn't take these claims seriously.

I'm still not getting how the funding of trials that don't succeed is supposed to happen. Individual researchers don't have the money to fund a serious drug trial on their own. So even if someone put a prize of 100 trillion dollars for curing pancreatic cancer, and you had an idea for a drug that you think can easily do that, but you'd need 10M dollars to actually synthesize and trial it, how do you go from here to there? Sure, you could sell equity in your prize, but how do you convince people to buy it?

Whatever you do, you can't get around the fact that the people with medical knowledge need to convince the people who have the money to invest in them before any research can take place. There is simply no way to get rid of the pitching to bureaucrats step here, not even in the most idealized societies - research just needs too many resources for any one individual to control.

My point is this: assume a world where free market maximalists win and patents (for drugs) get abolished, in the hope of reduced prices from increased competition. Since the reasoning is free market maximalism, this society will almost certainly not force companies to publish their drug formulas, it will allow them to keep these formulae a secret.

My contention is that in this case, the hoped for market competition would not materialize, drug prices would stay roughly the same for novel medicine normally covered by patents today. I admit that this may only be true for some drugs, not for all - some drugs may indeed be very easy to reverse engineer, and those would benefit from such a change. For example, the use of lithium to treat bipolar disorder would probably be easily reverse engineered and transformed and generic alternatives would quickly appear.

What I was pointing out with the generics part is that the formula publication requirement of patents is in practice having a significant positive impact today - all drugs that are off patent are being successfully synthesized and sold much more cheaply by generics drug manufacturers. This only happens for drugs that are far too old, so I'm not a fan of this system, but I think there is a real risk that even this small thing will get lost if we replaced patents with trade secrets.

> Since the people with the money don't understand the science, these possibilities must then be assessed by bureaucrats, and this causes our best to spend half of their time writing proposals instead of working and researching. A complete waste of time.

What part of this changes when a company is paying for the research, and why can't that same part be used in a government office? It's not like drug company CEOs and CFOs are also world class medical researchers. And if the funding decisions are entirely done by a CTO who is, then why would this model not be possible in a government office?

Ultimately someone needs to pay for all of the trials that don't work, and someone needs to do a cost/benefit/likelihood analysis on any research to decide whether it will be funded or not. With a public funding scheme, the benefit analysis could be based on public health interests. With a private company funding scheme, it has to be based on profitability. In healthcare, there is usually a huge gap between these two, unlike other more traditional markets.

The recipe for Coca Cola, the most popular soft drink in the world by far, has been a trade secret for over 100 years. And I'd wager a guess that it's slightly simpler to make then most medicines. So I don't see any reason whatsoever to assume that drugs would be easily reverse engineered.

Instead, I'd say that the assertion that today's drug making is hard to reverse engineer is hard even knowing the exact formula, substance, and dosage of every drug is unlikely to be true - is there any example of any drug that generics companies are not producing more cheaply than the original inventor, once the patent has expired?

I'm not really happy with the current system either - I tend to think that the state should just mandate that all medical research be fully published, and either have companies compete on research and manufacturing, or pay for it directly if it can't be made to work as a private business. My point was just that this is very much not a capitalist or libertarian position.

> If the US was maximally capitalist it would be a free for all with no patent protection.

The much more likely alternative in a maximally capitalist / free market maximalist society would be keeping all drug formulas as trade secrets, and thereby having all drugs as branded, no generics whatsoever (or few - perhaps some substances could be reverse engineered). In such a society, having the state force companies to publish their formulas would be seen as unacceptable interference in the free market, almost certainly.

The maximum libertarian alternative to patents isn't free-for-all copying, it's trade secret formulas - e.g. Coca Cola. Drug patents actually exist as a compromise given the clear need for the state to force companies to publish their drug formulas for research. Allowing companies to just keep their drugs secret would be even more capitalistic, and would increase drug prices even more.

If you are running Emacs, either in the terminal or in graphical mode, pressing Ctrl+C does not generate any interrupt signal, it simply passes the C-c key to Emacs, which is just a general key bind.

Now, C-c in particular is the key stroke which non-core modes are recommended to use for their key binds (for example, sgml-mode uses C-c as a prefix for all of its commands, e.g. C-c C-f for sgml-skip-tag-forward, C-c C-b for sgml-skip-tag-backward, etc). Running `killall -SIGINT emacs` will not be interpreted as a C-c key being hit in emacs, it will just cause emacs to quit.

The same thing is true for other terminal based programs, like vi. Pressing Ctrl+C in vi does not generate an interrupt signal, it just passes this key combo to vi to do with as it pleases.

Ctrl-C in Emacs is not "terminate program", it is "start of user command", in most modes. Similarly, even in vi/vim, Ctrl-C does something completely different. So this has nothing to do with the terminal whatsoever.

Most proponents of the existence of qualia regard them as fundamental to what they call the hard problem of consciousness. People like Chalmers and Searle assume that all of the externally observable behaviors of people could be (at least in principle) explained without the need for qualia; that is, they believe it's possible for p-zombies to exist; this then gives rise to this question of consciousness - if all physical behaviors can be explained without qualia, and yet everyone of us has qualia, how and why does this happen?

I thought the point above was in a similar vein - that qualia and language are theoretically separable phenomena, so that we can imagine a being might have qualia without language, or language without qualia, and so these need to be explained separately. I was trying to point out that we have no proof for the existence of beings that possess these two qualities separately, so that I don't think the theoretical distinction is necessarily true. Just like any volume of gas has a temperature and a pressure, the existence of separate concepts doesn't mean they are physically separable.

Modern Hebrew is built on Biblical Hebrew, on Mishnaic Hebrew, on Medieval Hebrew, on Yiddish, and has influence from many other languages. However, there is nothing similar to how English and other native languages evolved. Between the 2nd or 3rd centuries CE and the 20th century, there were no native monolingual speakers of Hebrew - even Mishnaic Hebrew was used only as a second language; while Medieval Hebrew was a lingua franca that Jewish populations speaking other languages would use for communication, not a native language to any of them.

Also, while the creation of Modern Hebrew began around the 1880s, it was an extremely niche phenomenon until much later - at least according to Wikipedia, in 1900 there were fewer than 10 families even in Ottoman Palestine speaking Modern Hebrew currently.

> It’s far more likely than not that they do experience things.

I don't think this is a fair statement. I also happen to believe that there are no gods and that most animals have subjective experiences in some sense similar to our own (while I also believe that atoms and rocks don't), but I don't really think there is a meaningful probability you can associate with either position.

This is pure pseudo-science at its finest. You're using sciencey words without putting any real meaning behind them. A frequency has to be a frequency of something (e.g. an electromagnetic wave). There would also have to be a mechanism that would explain why each body is tied to exactly one and only one "frequency" throughout its entire life, and why it never happens that different bodies share a consciousness; and there are thousand other things - but I'll stop here as I've already put more thought into this non-idea than it deserves.

> The point is that bats do have a subjective experience of the world which is very different from a person's. It seems like you think only humans have this?

This is definitely a possibility given the very basic level of understanding we have of this. The reality is that we don't know, and we don't even have a well defined way to know (that is, we don't even have any idea what kind of proof we would need to bring that animals have an experience of the world in some sense that is the same as ours but different from a rock's).

> And a modern Hebrew speaker can easily understand Biblical Hebrew — they're closer than even Modern English and Shakespearean.

Of course, because modern Hebrew was constructed based on (the modern understanding of) Biblical Hebrew around the 1920s or slightly earlier, whereas Modern English naturally evolved for ~400 years from Shakespearean English and other forms of English.

"American" would be ambiguous in this context, right? Both the Cherokee and the English-speaking residents of the USA are American, but the specific point here was about whether Sequoyah was only a Cherokee speaker or if he had any knowledge of English, Spanish, French, Latin or any other European language. In this context, saying that Sequoyah's father was "American" would not make any point - it wouldn't tell anyone reading the comment that he was not Cherokee nor from any other native population; whereas European makes that point succinctly.

> qualia totally precede language; one could experience qualia without language, etc.

While I do believe this as well, I don't think there is any way to prove this with current knowledge. You can introspect and separate your experience of a color from your language, but this type of introspection can also be misleading. And that's about all you can do - we don't know of any way to objectively test if another organism experiences qualia, and any historical/evolutionary evidence is also lost.

UB has 2 very different implications:

1. It means that even if your program happens to work, it can't be portable

2. It means that even if your program happens to work today, it might stop working tomorrow when you add some new code, when you change some compiler flags, or when you do even a minor compiler upgrade

Of course, a compiler can't address 1. However, a compiler can very much address item 2. If Microsoft were to say "in MSVC, we define integer overflow to wrap", then they would guarantee that `INT_MAX + 1` will produce `INT_MIN` regardless of any optimization settings, any compiler upgrades, any other changes to the code. Of course, compiling the exact same program with Clang or GCC might cause it to crash or corrupt memory or anything else - but as long as you stuck with MSVC, your program would have perfectly defined semantics.

This is similar to using compiler extensions or intrinsics - they are not portable and not defined by the standard, maybe even explicitly defined to NOT be supported per the standard (such as variable length arrays in C++ in GCC), but they are nevertheless perfectly safe as long as you stick to your chosen compiler.

Edit to add: the integer overflow example is not just a theoretical possibility - lots of C++ compilers provide the `-fwrapv` flag; when using that flag, signed integer overflow is no longer UB for that program, it is defined just the same as unsigned integer overflow.

Note that, while complex, there exist GCs that can handle both soft real time and even hard real time constraints - especially for Java. Memory overhead is a problem with GC languages, though, and that one is by design.

> No - if something is UB in the spec, it's UB.

A compiler is still free to ignore the spec and declare that something is not UB. However, this is very much compiler based, not platform based. Windows might guarantee that aligned DWORD-sized memory accesses are atomic, but that doesn't mean Clang when compiling for Windows would respect this - but MSVC might.