I know this is beside the point but I'm quite amused by this statement. Are you saying you're a totalitarian? I'm not trying to poke at you here; I'm genuinely interested what you consider the furthest thing from an anarchist to be?
From an Anarchist perspective isn't any state necessarily totalitarian? But I do believe there aught to be a strong, patriarchal, centralized state with strict laws and severe punishments. I think Singapore is close to an ideal.
Absolutely: missing in-person learning due to COVID. Less attention span due to growing up in a distracting environment. A lower bar to entry due to removal of standardized testing and indirectly from No Child Left Behind. Changes in parent or student attitudes. It could be any number of things, and it's lazy to just say "with AI usage" as something that has increased at the same time.
The understanding is inside of the system, in LLMs and in the Chinese Room. I agree with Daniel Dennett that it's preposterous to say that Chinese is not understood in any meaningful sense in the Chinese Room scenario -- it's just that the understanding has been hidden away in the background of the scenario.
Language is tremendously complicated. "Time flies like an arrow, but fruit flies like a banana." "Hard hats must be worn on site; dogs must be carried on escalators", etc. Predicting the next token requires understanding, full stop.
> if the rules are followed, no understanding is neccessary.
Autocomplete can predict the next token without understanding. It's a matter of probability. LLM's predict the next token with much more accuracy because they have much more data on token spatial relationships. To me that's not "understanding", that's "impressive statistical pattern discovery". The difference between a machine and a human though is we can infer that pattern discovery from a small data set, whereas LLM's can only do that because they have the entire internet's worth of data and then some. Of course you're going to make pretty good predictions of text if you have all the knowledge humanity has ever created. I'd be more impressed if you could make a small language model that could reason from first principles to become a large language model.
It is clear that consciousness is independent of the substrate if you don't believe in magic.
We could make a very dumb biological calculator out of a few genetically-engineered neurons that would very obviously not be conscious.
It's still an open question if we can embed consciousness in our current microchips if we had enough of them together (which I think we currently don't), or if it requires some other physical process we don't fully understand, e.g. quantum. I strongly doubt it does require any quantum shenanigans, but even if it did, we can and will find all sorts of ways to make computers that can perform those shenanigans too. Eventually we're just going to stop being able to move the goalposts, unless you set those goalposts in magic-land.
Quantum physics isn’t shenanigans, it’s completely fundamental and necessary to the brain and its electrochemistry. To say “yes but we can create a model of the brain which doesn’t need all that, we can model the classical physics or a reduced set of quantum interactions (running classically)” and then model the brains neurons on a computational substrate, well this is going to be nothing like the brain so won’t give you the result you’re looking for.
And say what you want about meat but we don’t seem to find consciousness in rocks or plants or clouds or hairdryers. And the buddists report that some very strange things happen if you meditate for years on end but obviously they must be talking shit and making it up because it’s not testably scientific.
Why always the strawman about "magic" or "supernatural" in these debates? The universe can be natural without it being fundamentally physical or functional. We don't know that consciousness is functional, in fact one could argue consciousness is the one thing that isn't captured by functional explanations. Because when you explain how a brain functions, the experiential part is left out and just assumed since we have brains that have experiences. So obviously there is a correlation. But we don't know that this correlation exists for anything functionally equivalent. We don't know that consciousness can be simulated.
My view is physical, functional, mathematical and informational explanations are abstractions from shared experiences. Abstractions remove the experiential aspect to arrive at something objective so we can understand the world around us, since experiences are creature and to some extent, individually dependent. This is Nagel's ultimate point in What It's Like to Be A Bat paper about the objective/subjective divide. And probably related to Kant's argument about phenomena vs the noumena outside experience. We try to understand reality via abstraction.
And it has nothing to do with magic. It's an epistemological situation we find ourselves in, which may or may not tell us something fundamental about the world. Depends on your metaphysical assumptions as to what can fundamentally exist.
> Are you conforming/obeying when you believe the Earth is round? That the sky is blue?
No, I am incorporating multiple different lines of evidence from multiple sources, including my eyes, into a framework of knowledge that I am constantly challenging and questioning, and "the Earth is round" and "the sky is blue" have survived those challenges as good first approximations to the truth. Whereas "Jews control the world" has extremely flimsy evidence, strong counter-evidence, doesn't fit with my understanding of the world, and can be traced as a myth/meme to known bad-faith actors. Which, by the way, is all also true for "vaccines cause autism" and "the earth is flat".
> inhibit the police's ability to arrest drunk drivers
They have breathalyzers and blood tests. Field sobriety tests are not there to help police arrest drunk drivers, they're there to help police arrest whomever they want to.
> I wish we would focus on the actual crimes the police are there to stop as much as we do the police reform.
The U.S. is one of the most punishment-happy countries in the world. Nearly every politician vows to be "tough on crime". This is an incredible thing to say given the past 50 years of policing and justice in the U.S. Won't somebody please think of the children!?
> I am neither left nor right
The "center" is constantly moving and has been, on average, shifting far to the right over the last 20 years. Anyone who claims to be a centrist is therefore either changing their politics with the wind, or was far right all along.
> They have breathalyzers and blood tests. Field sobriety tests are not there to help police arrest drunk drivers, they're there to help police arrest whomever they want to.
You're wrong about that. "Sobriety" isn't limited to alcohol. You'll notice that most laws against drunk driving are actually against being "intoxicated" or "impaired". Breathalyzers and blood tests are for gathering indisputable evidence.
Field sobriety tests are there to determine if you're motor skills are impaired. If an officer observers a person driving erratically and they can't walk a straight line or touch their own nose, they shouldn't be driving. You can be arrested for DUI [of sleeping pills].
The only time police would specify a DUI was for alcohol is if a breathalyzer or blood test showed that. Even if the officer says there was a beer can on the floor and they smelled like alcohol, they could be under the legal limit and be on any number of other substances so the DUI wouldn't specify alcohol.
> If an officer observers a person driving erratically and they can't walk a straight line or touch their own nose, they shouldn't be driving.
There are plenty of reasons that someone might not be able to demonstrate this to the subjective opinion of an officer and be completely unimpaired and competent at driving. e.g. people with atypical minds or bodies
Police generally ask people to do these tests when they have already made up their mind about someone being impaired. The only point of the test, practically, is generate standardized documentation. It is a dog and pony show.
Other countries that have serious anti-driving-impairment programs don't use these types of subjective tests -- they test people for using the substances directly.
They'll have an opportunity to prove that in court. I know that's not a great solution (because of the penalties involved in simply being accused of a crime, but that's a different issue) but, remember, they were pulled over for driving erratically and the, through conversation, the officer would gain further reason to ask them to do the test. The problem is the driving, everything after that is evidence gathering.
These days, so much of that will be recorded on video, from the dash cam to the body cam, it's usually cut and dry that the person accused is under the influence of something.
> people with atypical minds or bodies
This is a reasonable concern so I don't want to dismiss it but this isn't even close to the typical situation and, to emphasize, the reason for the stop is usually bad driving and the officer is looking for an explanation. Before a sobriety test is administered, there is already a cause for being pulled over. So people who can't pass a sobriety test because they have a physical or mental reason they can't only have that one piece of evidence against them removed.
I'm sure you can construct a hypothetical case where a person with a speech impairment, an inner ear deformaty and who's eyes shake when moving left and right gets arrested for DUI because they appear impaired but they weren't pulled over for those reasons.
The problem is that low-quality evidence causes both type 1 and type 2 errors.
Not only does it cause significant problems for people who are unjustly jailed and charged for crimes they didn't commit -- but it also lets drunk drivers off the hook when the flimsy evidence fails to convict. These aren't hypotheticals, both are very common.
Police in the US simply need to be equipped with roadside chemical tests for substances. They exist, they just simply don't use them.
> The officer takes a sample of your saliva by placing an absorbent collector in the mouth or on the tongue. The sample is then analysed at the roadside. If the test is positive, it must be confirmed by laboratory testing before charges can be laid.
Doesn't that sound like a better solution than: "The officer makes you stand on one leg and say the alphabet backwards, if they don't like they way you did it, you are charged with DUI"?
> I'm sure you can construct a hypothetical case where a person with a speech impairment, an inner ear deformaty and who's eyes shake when moving left and right gets arrested for DUI because they appear impaired but they weren't pulled over for those reasons.
The more common, and even more scary issue, is that sometimes people undergoing medical emergencies are arrested for DUI and sent to jail instead of a hospital. Which is again another situation that would be avoided entirely by roadside testing. This is such a common issue for diabetics that police normally do train to recognize the difference, but since they are not medical professionals and don't have adequate equipment, they still often confuse the two.
> Police in the US simply need to be equipped with roadside chemical tests for substances. They exist, they just simply don't use them.
I feel like you're not getting my point.
> Doesn't that sound like a better solution than: "The officer makes you stand on one leg and say the alphabet backwards, if they don't like they way you did it, you are charged with DUI"?
No, it doesn't. DUI isn't a law that lists a bunch of chemicals that are illegal to drive while using. The purpose of the tests is to prove you shouldn't be driving, not what drugs you're on.
You could be over-tired and get a DUI and I think that's justified.
Are you sure about that? Maybe your locale is different, but the DUI/OVI statute in the part of the US where I live is for influence of "alcohol or drugs" specifically. And it absolutely does have a list of what qualifies... and that list basically includes all drugs and alcohol.
Part of the criminal element of DUI is someone's choice to alter their body intentionally. I don't think many would think it would justly apply to natural processes like sleepiness or medical emergencies. That's not to say there shouldn't be penalties for failure to operate a vehicle safely, but those situations are clearly very different than DUI.
What I said there was confusing. I was saying that _I_ was fine with an over-tired person being charged with a DUI as I feel it's just as dangerous and just as elective. It is not the law here. That was unclear, sorry.
The relevant section of the law here is:
(f) It is unlawful for a person who is under the influence of any drug to drive a vehicle.
> And it absolutely does have a list of what qualifies... and that list basically includes all drugs and alcohol.
Does it state that the list is the exhaustive list? I have a hard time believing that someone could be pulled over, obviously inebriated but get off because they were actually on some new synthetic marijuana that wasn't on that list.
I disagree with that, because the penalties for DUI aren't merely because it is dangerous, but primarily because of the negligent intent related to the act.
And most laws in the US work this way (e.g. they are not strict liability). We often treat people differently based on why a thing happened because doing a bad thing for a bad reason is worse than doing the same thing for any other reason. For example, the penalty for killing a person might range from "absolutely nothing" to "life in prison or death" depending on why it happened.
> Does it state that the list is the exhaustive list? I have a hard time believing that someone could be pulled over, obviously inebriated but get off because they were actually on some new synthetic marijuana that wasn't on that list.
They thought of this -- the list itself includes any "chemical" that is "mind altering", as well as specific drugs.
What it doesn't include is any reason that isn't drugs or alcohol related.
> remember, they were pulled over for driving erratically
Maybe. Or they were pulled over for being black, or having tattoos, or being really hot, or because they criticized police brutality on social media, or because the officer needs to hit their arrest quota by the end of the month, or because they're driving an expensive car and the officer thinks they'll have lots of cash they can legally rob via civil asset forfeiture. We have far, far too many examples of all of these happening to say with any certainty that the police officer actually suspects anyone of an actual crime.
By the way, I have called in drivers who were badly impaired before. One kept driving up onto the curb, on the sidewalk and grass (next to a school!), then swerving back nearly into the oncoming lane, then stopping in the middle of the road, etc. Another kept swerving toward the concrete barriers on the highway, and when I passed them, they looked visibly asleep. Both times, the cops didn't care. They didn't send anyone. They sounded annoyed that I was bothering about that crap. The police do not care whether people are driving drunk or not, just like they do not care whether an active shooter is gunning down kids in an elementary school. They don't care if a violent dad with a restraining order has kidnapped his kids and is about to murder them, even when the mom tells them exactly where he has taken them. Their interests are orthogonal to the just enforcement of the law.
The final output of the neural network part of an LLM is a vector with weights for every token, that is then usually softmaxed and picked from. Can we not quantify the uncertainty by looking at the distribution of weights of the top 10 options? Like we expect for a note-taking app that the top choice would be something like 98% certain, and if we see that the model gives a weight of 60% to "Russia" and 30% to "France", that's just not enough certainty to simply output "Russia". That's exactly when it should say "<uncertain>" or something instead.
I’ve looked at confidence outputs for the chosen words from several STT providers and it’s definitely so that low confidence indicate that there is a risk that it has misheard.
Not always though. Let’s say that someone is saying ”1 2 3 4 <unintelligible> 6 7 8” then it will happily write 5 in the middle and give it good confidence as based on the context, it is the only likely word. Varies between TTS providers though.
Basically, why they are so good in average is that they estimate what is said most often based on the context. The context being then not only the audio but what was transcribed previously.
And if you don’t want it to be based on what is most likely to be said in context and only based on the audio around 1 word it is going to be awfully wrong most of the time.
It seems like the problem in this application is that attention itself. Makes me wonder if using a transformer for transcription is the correct architecture.
I mean, what I describe absolutely does exist, that's how LLMs work. The question is whether the relative weights are actually a good measure of confidence, and as the other reply to my comment points out, there are examples where it's not -- at least not the kind of "confidence" we really want.
I think it might break the game.
Most words sound similar enough to other words. "cat" and "get", "he simply" and "his simply", etc.
Add accents, and half the words would be indistinguishable from each other (note that word "indistinguishable", ironically, would be quite distinguishable).
People parse things like that in so much context, based in their own understanding of a situation, their grasp on speakers accent or speech impairments, etc.
Add to that that most native english speakers blur words together. The pause that in some languages is used to separate words, is used in english to separate sentences. English language as spoken doesn't separate words natively.
The text-to-speech before LLMs was meh. I think it's the ability to generate filler for uncertain words that makes it feel magic compared to before.
I think you're conflating opinions about when math is useful with opinions on the nature of math itself. Formalism does not assume that "all rules are equally valid". You can be a staunch formalist and yet still believe that X set of axioms are the only useful ones and everyone who assumes different axioms is wasting their time. You could be a formalist and still believe that the concept of infinity is leading math astray from useful math. Many of the differences you lay out seem to just be in people's opinion on which axioms are useful and which aren't. That's still formalism.
Setting that aside, it's very difficult for me to take non-formalist views of mathematics seriously. I strongly suspect that anyone who subscribes to those views has some deep-seated confusion in their heads.
> Platonism, which believes [mathematical objects] exist in some timeless realm beyond this physical universe
This is equivalent to formalism, except perhaps in how the mathematician feels about it. What could any possible difference be? In what way could it ever matter in the slightest whether something "really exists", if we define that to be so weak as to include "in some timeless realm beyond this universe"? Surely pink goblins "really exist" in this sense as well. With such a weak definition, the difference between your "really exists" and my "really exists" is purely emotional.
> Yet another view is conceptualism-mathematical objects really exist, but in the human mind.
You can be formalist and still argue about whether humans invented or discovered math. Beyond that, this is again just relying on the weakest possible definition of "really exists", with some added human-centric arrogance added in. Crows can count to 5; it's patently absurd to claim they are using something that is "not mathematics" or some completely alien form of mathematics that humans cannot access, because it's crow-brain math rather than human-brain math. This sounds like the Copenhangen Interpretation but for math: humans brains are magic! What are we doing? What are we talking about?
> This idea that some mathematical objects are in a philosophical sense “more real” than others is a big motivator of mathematical constructivism
Yet again, this is still formalism. Up until here, you've used the word "real" in such a weak tautological sense as to have no connection to our (or any possible) universe. But here, you've switched back to "real" meaning "having any bearing on our universe". So you're saying "constructivists consider different axioms useful than ZFC mathematicians do." More often they don't even really think about usefuless at all, it's just something that caught their interest and they decided to explore it.
> I think you're conflating opinions about when math is useful with opinions on the nature of math itself. Formalism does not assume that "all rules are equally valid"
I think you're misinterpreting what I was saying. Of course, a formalist will say that some rules are "more valid" in the sense that they produce more interesting or useful theorems. My point was, to a formalist, there is nothing more to be said about the validity of axioms than the value of the theorems they produce. Whereas, from certain other perspectives in the philosophy of mathematics, that is not the only grounds on which axioms can be judged.
> This is equivalent to formalism, except perhaps in how the mathematician feels about it. What could any possible difference be? In what way could it ever matter in the slightest whether something "really exists", if we define that to be so weak as to include "in some timeless realm beyond this universe"? Surely pink goblins "really exist" in this sense as well. With such a weak definition, the difference between your "really exists" and my "really exists" is purely emotional.
You sound like a logical positivist. And that's the issue – if your philosophical assumptions are positivist, then non-positivist philosophies of mathematics (and of anything else) simply aren't going to be intelligible to you. They can only make sense if you are at least willing to doubt for a moment your positivist assumptions.
> Crows can count to 5; it's patently absurd to claim they are using something that is "not mathematics" or some completely alien form of mathematics that humans cannot access, because it's crow-brain math rather than human-brain math. This sounds like the Copenhangen Interpretation but for math: humans brains are magic! What are we doing? What are we talking about?
Conceptualism claims that mathematics exists in the mind–but it doesn't claim necessarily only human minds. If animals have minds too, then mathematics can exist in animal minds as well, even if in a much more rudimentary form. I doubt any conceptualist would say, that if intelligent extraterrestrial life were discovered to exist, that their minds wouldn't contain mathematics simply because they are a different species from homo sapiens.
> So you're saying "constructivists consider different axioms useful than ZFC mathematicians do." More often they don't even really think about usefuless at all, it's just something that caught their interest and they decided to explore it.
There are different types of constructivists: (a) those who have a philosophical commitment to constructivism; (b) those who are interested in constructivism for practical reasons (related to computer science); (c) those who are just interested in it as an interesting mathematical system to explore. You can be (b) or (c) without needing any philosophical commitments at all, and they are completely compatible with a formalist philosophy of mathematics. And, quite possibly, the majority working in constructive mathematics today are (b) or (c) not (a). But, historically, the founders of constructive mathematics (e.g. Brouwer) were very much (a) not (b) or (c).
> There simply is no "non-formalist" mathematics.
I think you are conflating mathematics with the philosophy of mathematics – they are two distinct disciplines. Disagreements about the philosophy of mathematics make no direct difference to mathematics itself; at the margins, they can influence judgements about which problems are interesting – although, even there, a person can find ultrafinitist mathematics interesting without needing any philosophical commitment to an ultrafinitist philosophy of mathematics.
If this doesn't happen, are you going to accept that you were wrong, or are you going to ignore it and be off spreading unfounded anger about some other imagined offense?
No, the result of a calculation could be a key value or list or other compound value - whatever the result is. I am getting hung up on deceptive naming. If you have a 'result', the calculation is done. You have a result.
I know this is beside the point but I'm quite amused by this statement. Are you saying you're a totalitarian? I'm not trying to poke at you here; I'm genuinely interested what you consider the furthest thing from an anarchist to be?
reply