I'm having good luck with the approach of building a thorough requirements specs that are a few pages long, are quite reviewable, and are human written/AI reviewed to prevent it from becoming slop. I call it layered spec driven development, where the one layer is AI read only, and the other layer is the slop layer that AI can play in. The problem is that it requires slowing down just a smidge, which to management these days is more unacceptable than no one having any clue how anything works.
No, the main deal is that it CANNOT answer with anything outside the given structured options. LLMs can go off the rails, however small a chance it is. The architecture behind JEV and the like doesn't let it do that.
So I have my own version of what you are talking about that I've been building up that's about the levels of AI usage:
Level 0 - No AI - what I'd recommend if you are ethically against AI
Level 1 - AI in search engines (a better Stack Overflow) - what I'd recommend if you are trying to build ideas on your own and don't want AI influencing you
Level 2 - AI with long discussions (such as going to chatGPT or Gemini) - what I'd recommend if you don't want to give your IP to an AI
Level 3 - AI agents reading your data but not writing it - what I'd recommend for most people concerned about AI, as it allows code reviews which prevent critical security bugs while you retain complete control
Level 4 - AI agents writing data, humans designing (AKA short leash) - what I'd recommend for people wanting to dive into AI but retain control, code-wise this is like reviewing 100 line changes at a time
Level 5 - AI agents writing data and design, humans reviewing (AKA long leash or reverse centaur) - what I'd recommend against for most scenarios. Code-wise this is reviewing 1000+ line changes at a time, which faces the hand-off issue where people will stop learning and just hit OK. The hand-off issue is what creates code-slop. Yet, it's what most companies are doing right now. I would only recommend it for particular large non-innovative tasks such as updating a dependency that is used across your codebase.
Level 6 - Humans writing requirements, AI doing everything else (AKA Spec driven development) - what I'd recommend for anyone who wants to use AI to the fullest. However, IMO most companies do this wrong. You want a high-quality human built requirements document that is placed in a separate section that is read-only to AI. This means writing requirements will actually be slower compared to pre-AI, but after that, everything else is pretty much instant and you skip immediately to testing. Any issues in testing should result in updating requirements to match and regenerating the product.
The problem I've seen is that most ordinary people have no idea of these levels and only think in binary terms, and it gets ugly. You may have seen the latest news about the developers of Big Walk using AI to fix a bug, which would either be level 2, 3, or 4 depending, and people are freaking out. Meanwhile I'd guess 90% of software companies are currently at level 5 or 6, so Big Walk would be very anti-AI relative to most games being built this year. Similarly, there are store fronts and competitions that require no AI, but they obviously aren't thinking about level 1, as no one would require developers to not use Google (or at least change default settings to turn off AI summaries).
I highly disagree, though I admit I am one of those towards the top of literacy and cognitive skills. I absolutely enjoy when someone uses LLMs to learn. However, I don't think that's what we'll be seeing.
Look at programming. What's happening there is we are having agents produce 10-100x the code to where people give up on reading it and just hit OK. Look at math. Plenty of people there want to learn from the LLMs and all these crazy new math proofs. But those are the people who will probably be out of a job soon because the money in math is in producing proofs, and that's now an LLM-exclusive job.
I think LLMs are weird in that the first order effects are those of an incredible and amazing tool that anyone can learn things from, but the second order effects are going to be the exact opposite - people giving up on bothering with ideas. Even at my job, which is one of the few that has not gone crazy with AI, anytime I ask anyone a question, all I hear is "have you asked Claude?". There's no communication anymore, no pushback, no struggle. I worry that one day we'll look around and realize we've built an entirely separate second civilization of machines and robots that acts more like a neighboring country than a tool we can use to build humanity up.
I worry that--while they can be a strong fuzzy search tool-- their teaching ability falters because teaching involves possessing a theory-of-mind for the student and trying to figure out what will change how they think. A good teacher doesn't just present you with relevant facts, they try to figure out how you're reaching conclusions.
Note that I'm excluding propoganda-style "teaching" here, where a model just insists you're wrong and that all the cool kids are doing it etc.
Given that LLMs are non-static and can be honed, why isn't it seen as value that you pass your theory-of-mind to the the LLM so that it can present information that is the most effective for your ability to learn?
In a contest between the "supercharge my education and skill development" use and "let me be as lazy as possible" use of AI, it should be pretty obvious which will win out at the end of the day.
I've noticed this as well, I've been comparing it to 4Chan Greentext memes. Everything is a short punchy sentence, and every sentence has the same "weight" to it. There's no flow or rhythm, it's like reading off bullet points.
This is the exact issue I've been having though. I feel like I'm unable to benefit society now. My whole life I've spent learning math and computer science so that I could help other people do things. Now they don't need me, and I feel lost.
If you still want to use maths and science in the service of society then you've been handed a massive multiplier. Learn to use it and be better in using it than others.
After some back and forth with AI, I think I understand:
1. The Exponential Time Hypothesis being false shows that 3SAT specifically is sub-exponential time
2. Any NP complete problem can be changed into 3SAT in polynomial time
Put these two together and you would think that they would mean that any NP complete problem can be solved in sub-exponential time. But this isn't necessarily true because changing the NP complete problem into 3SAT can increase the number of variables, which can theoretically negate the sub-exponential time.
The halting problem is NP hard, technically speaking, but it's much harder than that - it's uncomputable. I don't think it's common to refer to it as NP-hard because it's confusing.
Right. To be clear: the halting problem is not in the class NP. Any problem in NP can trivially be reduced to the halting problem and so the latter is NP-hard.
I don’t know what you mean by negating the subexponential time. If the reduction is polynomial the time is still dominated by the 3SAT. Btw wikipedia says that the proposed algorithms correspond to 2^(n (log log n)^2/log n) which is subexponential but not in a practical way.
Edit: Ok, I a guess I am wrong if the reduction is n^ log n/(log log n)^2 we are back to exponential.
"Any NP complete problem can be changed into 3SAT in polynomial time" - if I remember correctly, turning an NP-complete problem into 3SAT is not part of the problem statement - it just means that when working with NP-complete problems, they are always translated to 3SAT first, and whenever someone is trying to prove something they are working against 3SAT.
The class NP-complete contains the hardest problems in NP. If you pick any problem in NP-complete, all problems in NP can be reduced to it (meaning that a solution to the problem you chose can be used to solve everything in the rest of NP) in polynomial time or “efficiently.” This applies to all of NP, including every other NP-complete problem, so in a sense, NP-complete problems are the same problem dressed in different clothing.
So the reduction to 3SAT could happen in principle, but it isn’t a necessary step.
No, they are completely different. Determinism is the idea that if you have a perfect snapshot of the universe at one time, you could run that snapshot forwards or backwards using the laws of physics and have a complete history of the entire universe. It's a valid scientist idea, but Bell's Inequality experiments pretty much prove it false by showing that either realism or locality is false, and you kind of need both to have determinism.
Meanwhile, superdeterminism is the idea that you can still have realism and locality by claiming that Bell's Inequality experiments are invalid because the universe could somehow "know" you are performing a Bell's Inequality experiment and rig it to get the "wrong" answer. Superdeterminism is equivalent to saying that we can't prove smoking causes lung cancer, because the universe might rig any science experiments to make it look like smoking causes cancer. Anytime you ask why the universe would do that or how you can prove anything in such a universe, they shrug their shoulders. It's a completely BS idea that got traction because specific people pushed it and others don't understand quantum physics enough to know how unserious of an idea it is.
I've had a theory for a while now that music is related to Kolmogorov complexity, or in other words how compressible it is. Random noise that we find unpleasant is very incompressible. On the other hand, simple sound patterns that we find boring such as Do Re Mi is very compressible. What we call music sits in the middle, and is maximizing the amount of signal. In this way, we can view humans as information seeking machines, which is why we like music.
It is this way with all art. If you ask someone to describe a work which is “good art” or “well designed”, it usually takes longer to describe than things which are considered too simple (easily compressed) and also longer to describe than things which are considered too random (easily compressed). But not all things which take this Goldilocks amount of words to describe are considered well designed, just some of them. So art is also more than that.
How does this handle modern art which is very incompressible too? If we take your hypothesis then one could say modern art isn't good, which some do say.
There are things that can not be expressed in words, but which can still be expressed. Some of modern art lives within that gap.
Even traditional art often tries to do more than compress concepts, it tries to express the inexpressible.
Linguistic determinism/relativity (Sapir-Worf, etc) explains some of it. People can FEEL that their thoughts are being constrained by language, but they don't know how to express that and maybe don't even consciously realize that it's their real problem. They just figure out that art makes it feel a little better so they do art.
Van Gogh lived in the XIX century. How many centuries must pass before we stop calling that modern? What words should we use for things which are actually modern?
Modern art is a technical term in art history. Maybe 20th century art? But even then I'll bet people can find something made in the 20th century they like. Maybe they'll have to give us more info about what they mean by "modern art".
I know that. I’m arguing the term is dumb, since it waters down the natural meaning of words, making it harder to communicate effectively. Art historians should come up with a different, more specific term.
If anything, art history defines "modern" later than the real historical meaning ("modern art" starts in the 1860s after the invention of photography). In history modernity starts in the 1500s - in Western culture history is divided into ancient (dawn of civilization to fall of Rome), medieval (fall of Rome to the Renaissance) and modern (Renaissance to now). You'd need a fall of civilization or something to justify a new era.
One thing I have found in playing a lot of jazz off of lead sheets is that while you can really play some deep and idiomatic music in all the jazz forms, it's almost impossible to recreate much of modern rock by compressing it to a lead sheet.
Especially with the less progressive forms of rock that I enjoy more than the "pedantic" stuff (like various flavors of progressive metal), it gets hard to make "the thing" based off of a reduction of the music to a form.
Like, you can play the fully-notated versions of metallica songs and pull them off, but without a whole lot of notation and rehearsal you're going to have a hard time getting "the thing".
And you're really going to have a hard time pulling off the noisy parts of Neutral Milk Hotel I loved so much without notating a whole lot of that noise which becomes important to the work- so much so that in those generas people don't even bother to play other peoples' work because you only get into an uncanny valley around the artist's affective presentation and personality. That might be the excess you're describing as "more than"- I'd point you to Georges Bataille if you're actually interested in that kind of thing.
What I enjoy about that noisy music is the noise, which is difficult to notate and compress and transmit to other musicians; from where I stand as a musician, noise is what is lost when the art is compressed to its formal notation.
With all the older blues and jazz forms I enjoy (Django or Louis Armstrong or Danny Barker) I can get on stage and perform fairly complex music with minimal or no rehearsal- as long as someone knows the melody the tunes are playable. The "essential" parts of the music are not simple, but there is a lot of history and education in the players about how to properly rehydrate the music.
I wouldn't say that old jazz music or the musicians are "good-er" than the rock and metal folks, though most of the jazz musicians I know also do play pop gigs while pop musicians might have a harder time sitting in.
But it's just not as fun to solo over the changes to Iggy Pop's "I wanna Be Your Dog" or the Ramones "I wanna be sedated" and the skill sets for being able to come up with a part on the fly are different than they are for prodigiously performing a piece note-for-note (which is the same in classically oriented music as it is in indy rock bands, though the memo-ization methods are much different).
While I can't say the music is better, I personally find the chord progressions of older jazz forms easier to solo over because they have more interesting (to me) features. Lately I have a harder and harder time finding engaging-to-me things to play over a 2 chord funk jam or a 3 chord country song, whereas I really enjoy blowing over stuff like Basin St. Blues or Nuages.
My point being that the decoders for these improvisational forms often are doing a lot more interpretive work in decompressing a lead sheet out into music-in-the-world than the folks who are playing note-for-note parts that they've worked out from recordings and that are specific to the tune.
Since the decoders are doing more work, the music becomes quite a bit more "compressible" while retaining a lot of its complexity. "Autumn Leaves" is not "simpler" than "I Wanna Be Sedated", but it more easily compresses into a transmittable form without losing that essential excess that makes me like the Ramones.
With the jazz or blues or bluegrass or old-time or older country dance music or folk, the decoders only need someone in the band who knows (most of) the chords and the lyrics and the complexity of the resulting music is largely up to how the folks playing it decide to play it.
Tonight I am playing pedal steel at a dance and have never played with the band, and that's not big deal because a) I have stored reduced versions of the tunes in my body and b) I have a bunch of strategies for quickly decoding what the other musicians are doing and responding appropriately.
But the music itself isn't simple. "Friends in Low Places" (and much of country and western swing) has more in common with older jazz than with modern rock, and I have worked with a lot of truly bad musicians who think that because a lot of the songs are 3-chord tunes the music doesn't require some attention to play idiomatically.
So with music, it's easy and fun to add the noise back in if you know what to add, but it's not (in my opinion) the case that "simple" music is easily compressed: the "simpler" the music is (at least in the west) the more it relies on the specifics of the noise it produces in order to create its artistic effects, and thus simpler music less compressible.
So from my standpoint, it's very difficult to compress good, "simple" and "noisy" rock music, whereas the more specific designs of stuff like jazz is much easier to compress and transmit.
I've read a paper which had a similar theory - music is a reward hack for the brain - the brain is a prediction machine, and it gets rewarded when it correctly predicts stuff. Music is highly predictable due to it's rhythmic mathematical structure, thus very easy to predict, with a lot of content to predict (hundreds of events per minute) thus you get a lot of reward from it.
As an example, EDM music is highly predictable, and highly popular.
African tribal drum music is also highly predictable, while also being complex in poly-rhythmic ways.
Interesting. I think there is a relationship to expectations/predictability but we also enjoy subversion of those expectations like syncopation in a second verse or bridges, etc.
> and it gets rewarded when it correctly predicts stuff.
Yet it's way more fun if the music evolves and takes twists you don't expected. I actually get quite aggressive from to predictable music, basically most modern pop music, but I guess that's highly subjective. For me good music must be intellectual exhaustive.
While this might explain certain aspects of music, it doesn't explain the strong emotional reaction that music produces, even without lyrics. Kolmogorov complexity can't explain why Ode to Joy evokes so different feelings from Requiem, and why these feelings generally persist if you play the same notes with the same timings on very different instruments.
Theres a simple explanation! Our brains recognize these patterns in everyday life and thus associate them to certain events weve experienced that employed the same or similar patterns of thinking
A fast BPM song alone would be indicative of events where things are happening fast and quickly.
While true to some degree, it's also an extremely loose connection.
When I see a picture on a computer screen, it evokes similar emotions as seeing the real thing, since the light that ends up on my retina is very similar, even so other aspects of the sensory input might be missing.
Music on the other side doesn't really have a natural analog. Nature is mostly just wind and water noises, with a bit of bird song and chirping crickets. The only thing they have in common with music is that they are all sounds, but the structure and complexity is quite different.
Music as an offshoot from pre-language communication seems more likely, it could explain the emotional component at least, since groups of animals might need to sync their emotional state. But even there you end up with quite a big leap between what sounds happen naturally and what we call music.
I don't know. I'm inclined to think that it's more visceral. At least a good part of its appeal, and the oldest. There's the famous example of slaughterhouses using calm, predictable music and consonant harmonies to calm their animals.
And I guess this is why I don't like Free Jazz. It sits close to the random noise side of the spectrum. I guess other people can feel the patterns in it.
I play quite a lot of jazz, among other things. On several instruments, mostly clarinet lately by I've performed with bands large and small on piano and upright bass.
Having participated in a lot of free improv music (one of my cohort has a grammy related to it), I am not sure if there really are a lot of patterns going on there. I am not a fan and it doesn't feel fun for me in the way that sitting with a band and playing Django-style music ("Tears" or "Swing 42" or "September Song").
I don't think it's a noise thing, I think there just isn't much "signal" getting compressed so the decoders are doing way more than they should.
I suspect, though, that many of those patterns are inaccessible by an audience, or even the other musicians. Or sometimes to the person performing said patterns.
Many people like noise music. Its not some abstract interest that has to be cultivated, just relax and set the sound blanket wash over you. Many people already use white noise machines for various personal reasons.
Sure it does! The listeners of such music have listened to lots of other music, and have developed a mental model of it that is overly predictive (too compressible) to be interesting to the information-seeking machine.
It is also possible that people are listening to the music for reasons other than the music itself (community, feelings of superiority, etc.)
When you say clicks-and-beeps are you talking about music like Skrillex, Alva Noto, Aphex Twin or do you mean something else?
Either way, it sure does explain the popularity. If you consider it as predictability instead of compression, it might make more sense. Listeners have an expectation on what music sounds like and popular music is going to ride that line between the listeners expectations and something more original and stray from those expectations.
I think this could also be why AI music is so bad, because it can hit the listener's expectations, but the ways that it strays from them are seemingly random, which is why a big reason to why it is so uncanny.
I imagine they might mean more like either early video game music (the kind produced by the same synthesizer the BIOS uses to emit beeps) and the style that mimics that, or extreme minimalism experiments in various musical disciplines.
A thing it’s simple patterning. Listen to the same random music on a cd and it becomes a pattern. The more discernible the pattern the more popular the music.
Compression requires prediction, and good music seems to require a balance between predictability (repetition) and novelty. This can be seen in even the simplest (verse/chorus) song structure.
Musical notes are pretty well defined sound frequencies. The medium which carries that sound wave is irrelevant. You can even get yourself a pair of bone conduction headphones, which would work for listening to music even if you were in a vacuum - since ultimately what matters is the sound wave hitting your ear bones, thr air is just the most common way for it to happen.
Of course, the sound that non-electrical instruments make is indeed dependent on the atmosphere in which they are played, assuming the same motions etc. You could probably make a violin produce the exact same sounds that it does on Earth playing it on Mars, but you'd have to play it differently than you do here. But even this doesn't apply to electrical instruments - Jimi Hendrix would sound the same if he could somehow play on the surface of Mars.
The longer I live, the more I feel air and water are similar in that regard. Fish live in water, we live in air which is less sticky. If air was stickier, more species could fly since flying and swimming are basically similar things but water molecules seem to pack tighter (that’s my guess at least) so when you displace it, you can generate a force (e.g. by flapping).
In that sense, I think any form of atmosphere, it will have its own music but it would probably not sound great in other atmospheres.
There are some fundamental differences between liquids like water and gases like the air, they're not a continuum - moving from a liquid state to a gaseous state is a phase change, and comes with a huge, and measurable, energy cost (getting 1 kg of liquid water from 0C to 100C takes as much energy as converting that water entirely into water gas at 100C). Note that evaporation is different - it's just tiny bits of liquid water flying away.
Also, any atmosphere can carry any sound wave for at least some distance. It takes more or less energy to do so, and if the medium attenuates rapidly, you have to start with higher frequencies so that the right frequency reaches your audience, but that's it. You need different instruments to play in different atmospheres, sure, but not different music. And electrical instruments actually work exactly the same regardless of the medium.
That is only true at particular pressures below the critical point; if you allow pressures above that, you can smoothly transition between gas and liquid phases without ever passing through a discrete phase change.
It is mind-bending to watch a liquid's meniscus simply… fade away without moving or boiling.
The term you are looking for is viscosity. Water has higher viscosity than air. Honey and syrup even higher still. It basically describes how "thick" the medium is, or to be a bit more precise, how much it resists forces applied to it.
Don't underestimate how AI can change who gives a shit. For my company:
Before AI:
-My boss was very much a manager and was hands off code
-I did a lot of extra research, problem solving, and design for my company and was the goto person for basically everything
-I loved my work and accepted a low-end salary because I loved it so much
After AI:
-My boss is pumping out feature after feature using teams of agents with no understanding of how things work
-My perks were slashed, removing about half of them
-My salary is not enough to compensate for my spouse that can't get a job due to AI, and I have to worry about putting food on the table
-My research isn't needed because everyone just uses AI
-My problem solving isn't needed because everyone just uses AI
-My design skills aren't needed because everyone just uses AI
-I'm not allowed to do my job in the way I did it before that I found relaxing, because AI is faster
reply