Some additional thoughts/corrections: 1) One thing that I do need to mention is that, while I don't think AI composers will have an adverse effect on artists' ability to express themselves, there is a very real chance that it'll reduce the opportunity for session work. That's the direction the industry's been trending for a long time, though, way before AI came onto the scene, and I do think there'll still be some prestige attached to having real humans play the things, at least in some genres. Still, though, it will probably happen, and it'll suck if it does, but I don't know how much of it will actually be AI's fault. 2) Another point that I didn't really have time to get into is that, so far, no one's gotten close to an AI that can actually _innovate_ musically. JukeBox can imitate artists, but its knowledge of music is restricted to the ideas it's seen: It can't really come up with new ones except by accident. It has no artistic vision of its own, which again makes it poorly suited to replace actual human musicians. There's a reason new artists keep rising to prominence, so an AI trained exclusively to replicate past artists will inevitably produce stale results over time. That's not to say that all AI systems will have this problem forever: If we ever get Artificial General Intelligence I'd expect it to be able to solve that problem, but so far there's no indications I'm aware of that we're anywhere close to that. 3) Thanks again to Jordan! You should check out her channel: ua-cam.com/channels/1H1NWNTG2Xi3pt85ykVSHA.html
id be interested to see how the ai handles non western stuff, i imagine that it was trained on mostly if not all 12 edo music, maybe it could make some bohlen pierce pieces or something
@@bigweld4328 I have audio processing issues that music tends to exasperate, I often can't understand what's being sung so didn't notice a large difference in comprehensibility - Maybe less comprehensible than pop and closer to (but not reaching) the incomprehensibility of (English language) classical. I got uncanny valley vibes from a lot of it, though. Most of it felt... "off", and I'm much closer to the 'average listener' than most people who watch this sort of video. Not bad, just... Weird. I'd get used to it in time, of course.
@@bigweld4328 I could understand it at least a little, but it had some seriously harsh digital artifacts, including sounding like the recording was speeding up and slowing down randomly, like different pitches of background static and such.
Discovered Jukebox myself a few days ago and found something interesting - its re-rendition of "Yesterday" by the Beatles has live audience applause at the end. An AI learned a - that songs sometimes end with applause b - what applause sounds like c - when the applause should start and when it should end I'm super impressed honestly at the depth and detail of understanding that this AI has self-generated and can use.
It's worth remembering however that the AI has no idea what "applause" is, what "The Beatles" are, what "music" is, or anything else. All it know is that certain sounds tend to follow certain other sounds. Yes, the level of pattern recognition required for that is crazy impressive, but it's equally impressive just how much more there is to cognition.
@@MrAidanFrancis Exactly, all it learned is that many songs it heard end with a particular sound (applause) and so it puts that same sound at the end of what it generates.
I was thinking, "Was that a hidden track on the album, because I've never heard that. That doesn't seem like the lyrics Paul Simon would write...Oh he's joking. Thank G-d."
Really? I'm not trying to be pretentious about this or anything, it's just interesting to me... That barely even sounded like words to me, much less something that Simon and/or Garfunkel would actually release. The general tonality of the phrases was right, but I literally had to look up the lyrics to the song to even figure out what it was supposed to be saying.
@@bronsoncarder2491 How familiar were you with the original song? That probably makes a big difference. If you hadn't heard it much, your brain wouldn't fill in the gaps after realizing what the S&G AI cover actually was like it would were you more familiar with the original work.
When these discussions come up, I always remember what Rush said: "All this machinery making modern music can still be open-hearted. Not so fully-charted, it's really just a question of your honesty." Until it's intelligent, it's just a tool. And if it becomes intelligent, who is to say it's any less a musician than the rest of us?
A musician making an AI video that isn't either completely uptight and dismissive of AI's legitimacy? This is a breath of fresh air! Keep up the great work 12tone!
@@R4ndomNMBRS Fux (Johan Joseph Fux, to be specific). He wrote the person's name down. And it's the low quality of the algorithm that means he chose to do that. "Low quality algorithms produce poor results" is... Well it's the Scunthorp problem but for audio in this case.
It's also worth mentioning that musical composition was the first suggested application for computers outside of computing purely abstract mathematical algorithms - it was suggested by Ada Lovelace, who also wrote the first computer program.
I found Jukebox to be pretty interesting. As you mentioned, all of the words it produces are garbled and the song structure is spontaneous, so while I think if you positioned the music it generates as having been created by a human, another person might not immediately say that it was created by an AI, but they would clearly be able to tell that something was off about it. That said, I think there's a different way to view it that might be more helpful. Imagine that you want to write a radio-ready song. You might take some music that you've internalized, steal some pieces from one song, mix in a dash of another song, and come up with some ideas that way. But would you be able to hear the entire song in your head from start to finish? Of course not. You might hear certain parts, but not a fully-formed thing. You need to play with those ideas, iterate on them, edit them down to get there. That is, the music in your head is just a prototype. I view the music that Jukebox generates in the same way. It's a soup of ideas in their most raw form. And if you take away the garbledness and just look at it from an idea generation perspective, I think it could prove to be a pretty useful tool for musicians.
I was about to complain about how he doesn't mention OpenAI and how claims that he did that all by himself.. *hears the first sentence after the intro* Oh. I forgot I'm actually watching a decent YT channel.. glad those still exist, TY!
I suppose it is comforting thinking of it that way, but I don't think it is realistic. Eventually AI will compose music better than people. I'm guessing in about 10 years time most music will be composed by machines.
@@KyleMart doubt it. it could make stuff that nobody really cares like elavator music. for AI to come to a level it really composes, not repeats learned pattern you can push that number x10 or x100. still then, AI will be good for workcomposing but it wont write hits.
10:00 wow I’m late with the timestamp but wow. Thanks for having the guest and lol at changing the color while they spoke. The information was WAYYY over my head as well for specifics but it was explained in a good enough way that I could try at least and wrap my head around it. It’s getting a little dicey tho 😅😅😅😅😅😅😅
Some of the attempts it's made at classical music are pretty interesting - orchestral stuff in particular seems to work quite well. There is also a piece claiming to be in the style of Beethoven that's more like weird jazz fusion-ish... something, though, so I wouldn't be super worried about it overtaking human composers just yet.
Remember: it's not about the rules, it's about how to break them. The journey of musical AI is more or less the same as for humans and their way of getting creative: you start with learning; then you repeat other people's ideas (for example, you re-play your favourite guitar solo); the next step is to re-arrange and restructure other people's ideas, adding your own ideas more and more over time; later, your own ideas are so predominant that other people's influences are not as obvious anymore as at the start of your learning curve. This is still based on rules (musical traditions, listeners' expectations, expectations of your audience, etc.) and how to apply them. The real artistic challenge is to break the rules in a convincing manner - be different and be relatable at the same time. "relatable" means you follow the rules (at least in some way), so people can understand what you're doing and can follow you. "different" means you don't follow the rules, so you are unique (at least in some way) and your work stands out, compared to those you strictly follow the rules. So what does this mean for musical AI? The AI will become human if it knows how to and when to break rules. An understanding of 'breaking the rules' would turn an Artificial Intelligence into an Artistic Intelligence.
1. Program the rules 2. write a program that breaks each of them in turn (and then in variations) 3. have a another bot listen and evaluate and pass the interesting ones to a UA-cam channel and let humans vote.
The other thing worth pointing out is, the team at OpenAI is not trying to generate music. If they were, the music would be better. What they're trying to do is learn about intelligence and neural networks, and they're using music generation as a way to do that. Certain design choices that might result in better music would also make the system less useful from a research perspective.
With as much screen time given to them, you would think "The Sharpie" company (Newell Brands) would contact you with a deal of some kind. Anyway, excellent video and thanks to Jordan for the breakdown.
As i view, it could easily change the way background music for media is produce, but not affecting the whole industry. As you said, when someone follows an artist it's because that person shares a point of view, a culture and also empathize with that artist's feelings. It's not just sound, it's human feelings coded with sound.
I'm currently doing my Doctorate on AI Popular Music by investigating perceptions from musicians/non-musicians on creativity, performance, and musicianship as well as its capacity to form music scenes. This video was fantastically accurate and I'm glad to see more of this topic being covered by excellent UA-camrs!
Funny how similar it is to how a brain makes music intuitively. I had an instance of sleep paralysis once, and I hallucinated a classic rock music track that just went on and on with no real structure, while sounding constently good and real. I could even change the language of lyrics from Russian to English and back, and, while the lyrics had no strict meaning, they were made of human-sounding words. I've managed to remember a couple of lines, and they totally sound like AI lyrics.
The only thing I want out of AI music generation is, arguably, the most unethical. I want to be able to take one artist's vocal performance and transmogrify it into another artist's voice. I want to take, for example, Kurt Cobain's vocals off of "Where did You Sleep Last Night?" And change it into John Lennon's voice. We're almost there as it is, and vocal synthesis from scratch already exists.
@@GibusWearingMann oh yeah, I've been playing with photo and video style transfer for a couple years now. Audio is a whole different ballgame, because a normal sample rate means that it would have to generate ~40k "pictures" per second.
You'd be better off hiring a well-trained impressionist with a similar physical build (the body is an instrument). The best you'll get with AI is a pitch-corrected version. Adam Neely did a video where he "pitch corrected" Robert Plant, Aretha Franklin, and others; and the results were... not good. Furthermore, 13:57 in this video.
One possible way that might help with allowing for learning structure without putting the storage requirement out of range would be to convert the songs to MIDI, learn whole songs that way, then convert back to audio.
AI-generated music reminds me of "photo-bashing," a technique currently dominating concept art. It's a nifty bit of workflow in which artists skip 95% of their brushstrokes and use photo collage instead, to create visually-gorgeous 20-hour paintings in about 20 minutes. Super cost-effective, beautiful when done correctly, highly sought-after in film production. Also, limited to clinical realism, and results in mechanical designs that look like A) tennis shoes or B) oil rigs covered in tennis shoes. Whenever generative music does take off, it will have some _weird_ hallmarks. I can't wait.
I don't think AI in particular is a threat. I worked on AI for composing. The music industry has tried to rid itself of session musicians since its inception. Some computer has already generated all possible rhythyms and melodies. All the AI's do is use heuristics whether markov chains, random forests, or machine learning to pick a possible structure that it predicts will sound good. Whether it be in audio or notation it's unlikely that it will be able to replace humans, but it could certainly be used to create music for bands who onlh have starts and have some presence in the market. It's tricky to say exactly what will happen in a macro sense.
I am totally on the opposite side. AI will help (already helps) people understand what elements make music sounds nice, and give you new ideas on how to make your music stand out from other music (which will help you ironically, but very beautifully "personalize" your own music style).
@ghost mall even better is how ai can tell us new things about how we produce music and the hidden variables we dont necessarily notice conciously, since because ai uses our music as a basis, it takes part of this essence into its model.
At first, I was ready to believe that 'Shape of You' was a genuine Simon & Garfunkel song. The only point of suspicion was the dubiously modern title...
If it enables me to be more creative and do things I can't do now I would give it a try. I am always limited by the tools I have. However, doesn't mean I'll ever give up my instruments. Even if none of my music makes a chart list, I do it for myself and pleasure first and foremost.
There is an error in the description of digital audio. It's true that it consists of repeated samples, but the samples are describing sound pressure level (or amplitude), not frequency
While this is very, very impressive, and seeing technology progress so fast is somewhat scary, I can't help but feel like this is somewhat of a gimmick. With some of these AI models being around for a while, you'd expect musicians to actually, you know, use them. But we haven't really seen that. It only took a couple of years for autotune to be used on a number one hit song (Cher's "Believe"). Why hasn't that happened with AI-generated music yet? Generating a whole song is still not quite there, yes, but that doesn't change the fact that there isn't a top 40 hit that has had, say, a verse or chorus, or even a hook generated by AI. I think that calls into question whether AI will really take over, at least in the department of whole songs.
taking- over is one thing (probably not going to happen in any of our lifetimes) but since when is corporate music the dipstick of the avant- garde? very serious musicians do in fact make algorithmic music. acreil comes to mind.
I am now really curious to know what happens when you feed that Hotel California bit back into to JukeBox and have it try again based on it's own take.
My first intro to AI-generated music was Nick Didkovsky of Doctor Nerve (avant-prog-metal). I think a lot of his AI-generated work was based on Markov Chains, but I don't remember for sure.
One thing I'm concerned about is the fabrication of paratext, as it's already something done in the commercial world. I've been meaning to explore an art series of fabricating paratext in order to emphasis its role, but I lack the technical ability to execute such a project.
Think in the historic overview the "interactive phrase synthesizer" (IPS) introduced as part of Cubase v2 from the mid '80s deserves a honorably mention. It is more of a inspirational tool, but probably contributed to more pop and film music productions than all other algorithmic composing programs combined.
I've seen a few of these types of learning AIs done with stuff like Talk to Transformer or AI dungeon, and I've even used what seems like a midi version of this with a chord progression I wrote a while back, and more often than not it'd just...go off on its own, lol. And music aside, it always had trouble determining intent. AI Dungeon characters would do things for no reason and forget why they did it, and music pieces would drop what they were doing for no reason and switch to an entirely different genre, or ignore the test data. It's interesting to hear what it does with music, but I'm not scared of it by a long shot, lol
Should we be worried? Even Sudoku are much stronger if hand made than computer generated (Though we haven't seen anyone apply deep learning algorithms to sudoku yet, so... Unfair comparison, maybe), so I don't think there's much to worry about here. There will be a place for generative music - Projects that desire more music than is practical to actually create, in the same way that there's a place for generated levels - games that require more levels (for whatever reason) than is practical to actually create. What I'm less sure about is if deep learning AI will be better for those projects than 'build a bespoke algorithm that gives exactly the sort of music you're looking for' - Which is still how good level generation is done in roguelikes and other games with proc gen.
When advanced General Artficial Intelligence comes along, I do believe that it will make most of mainstream recording artists obsolete. I anticipate a subscription service that analyses your specific taste and generates music that's right down your alley. You can also give it manual input to influence the output, like how much of a deviation/surprise factor you would like to have in your music. There's probably also going to be an indie counter-culture of people that try to listen to only "real" music, but they can also be fooled by AI generated music that gets put online, as it is now indistinguishable from music that was recorded by humans. Also, they won't be able to resist the pull of generating new albums from their favorite bands that no longer put out music. That leaves live music, which might be the real last bastion of humanity. Then again, with robotics advanced enough, you could put together bands that are indistinguishable from humans, but never get tired or have an off night. They spit, they sweat, everything. I know, it's far off. But that doesn't mean that it can't become a reality.
14:08 people do not create a connection to the composer. They create a connection to an imagined caricature of the composer. Since that caricature has little to do with the real person there is little need for it to be an actual person. As long as the listener can create the caricature, Then the Human/AI status of the actual composer doesn't matter.
So overcoming the difficulty of it having to be trained with 24 second samples is a bigger issue than it seems. Broadly speaking, this is needed to do something called decorrelation. Due to reasons lending to probability and statistics I won't get into, if you give any current AI structure a sequence that's too long, it will start to memorize that sequence, breaking its generalization power, especially for elements later in the chain. What this means in practical terms is that likely if you trained the AI on whole songs after the first 30 seconds or so it wouldn't be able to generate anything coherent unless you specifically gave it the first 30 seconds of an existing song (at which point it would just regurgitate what it learned, but more than likely this would be mangled too). This means generally what you do in these cases is what's called "minibatches" where you artificially break up your longer sequence data into smaller subsequences. This decorrelates the data and produces better results, but as you see it breaks some of the structure. All told the fact it's over 20 seconds is impressive, just a couple years ago that would be *way* too long for any system to handle, but on a musical scale that's pretty short unless you're generating ad jingles or something.
Just for the fun, i checked Jukebox link, and looking for some metal examples i found out a Judas Priest example with lyrics about having tons of ganja in his head xD
I would be so interested to hear it make music in the style of Dilla's Beat 27, Tomorrow Never Knows, Get Dis Money, Spacey and many more of what easily have a place among my favorite tracks of all time.
Would love to hear Jukebox write some doom, sludge, or stoner rock. Would also love to make it compose a continuation to something like Nina Hagen's Naturtrane
Hey, what about people who make a living composing for movies, ads or video games ? For instance, seems very likely given an AI that can imitate the "Generic orchestral film music" style well enough, a lot of actors in these industries will gravitate toward that rather than the more expensive human composers. Of course there are a lot of great, truly unique video games or movie soundtrack, but It feels to me like transmission of an universal emotion from human to human is less important in that context ?
"like the transmission of a universal emotion from human to human is less important in that context?" it doesn't have to be. cost (monetary(monopoly money)) efficiency has come to supersede actual efficiency and the cost to implement versus the value lost from a humane score will be a determining factor in our march towards complete optimization of human activity
Well - but in the end, it's all our choice and taste. Also - there is still a need for programmer and someone who put data into the system. So I don't think that AI will ever write music by itself. I guess on the contrary there will be more creators - it give open door to more people to create music as it becomes easier.
@ghost mall Having grown up with Relient K, I'm just imagining it trying to fuse the heaviness of I Need You (a hard rock song showing the lead singer's dependency on and need for God) with the lyrical levity of In Love With the 80's (I figure that song name is self-explanatory), having no idea that its chosen genre and lyrical content don't exactly match...
Should musicians be worried??? NO Because even the Old Jukebox the literal jukebox WAS technically a threat to live performance of music. Music really is special.100 years ago if I knew that UA-cam and portable music will exist I would bet my life that no person will ever learn to play live instruments anymore and no one will be hired to play for restaurants etc since we can access music at whim But surprisingly musicians still exist
@1:37: Player pianos run on paper tape--it's not a big stretch to make a paper-tape computers punch their own tapes. @8:41 Still counts as Rick Astley's classic.
I think Ai music should be influenced by other artists like Jukebox is doing, but not try too hard to sound like the original artist. How do I get to do Jukebox? I want to try some things with it one day. Or do you know if you could do it. Since I am only 17 and don’t know if I could get it yet. Will it do the song I’m a little teapot in the style of Abubakar Sani?
If EMI was recognized as a the real Bach of the samples does that mean it was somehow able to get closer to "Bach-ness?" Like some platonic ideal of Bach's music? This stuff is wierd.
That's actually really interesting because I had a (somewhat) related experience a while ago. I was writing a song and did something in the style of Bastille (a pop band). I told my gf how it's a thing Bastille does, and she wasn't sure what I meant, so I went on the hunt for an example. Long story short, it took a LONG time to find a song where they do it, and I felt like I was crazy. It goes to show you that a "style" is really just the amalgamation of lots of things that the artist doesn't necessarily always do all at once. It's like how AI generated images often blur the edges, they're doing every part of an image at once (background, foreground, etc).
There has been prior research on generating caricatures of human faces by amplifying their position in "face space". So it wouldn't be surprising if you could generate music that is Bach-ier than Bach.
I would guess it had been trained on all of his work, in which case whatever it produced would be sort of like an average of everything he ever did. For the experiment to work, you'd have to pick something less well-known of his so that people don't recognize the actual piece, so it might be less representative than an "average", particularly since he had a long career and his style probably shifted over time.
If we don’t have income reliable enough to support you on patreon, but we still want to give a small one fine piece of financial support, is there a way to do that? Also, fascinating video! I’ve been playing a game called the Talos Principle which also deals with how we should treat AI, so this actually came at a very good time for me 😂
Sure it might affect the artists and theindustry for some time? But not forever, photography didn't make painting obsolete. We might need laws to protect visual and musical artists from AI developers. It's highly unethical that they went ahead and took artists work and fed their AI with it (like with Midjourney), without the artists permission. Imagine training your voice your entire life, honing your craft, then a developer takes your unique style, feeds it into an algorithm, so someone can type in "blah blah blah" and have what took you years, in just a few clicks. I know there are some technology lovers who think that this is a good idea, but most people would probably disagree. I can't really think of the excitment someone would have to listen to music written by a thing with no feelings. Cuz that's the thing in music, artists (who are people, like the listeners) can obviously undestand and express more. Just an example, Je te laisserai des mots by Patrick Watson is a song that affected so many people in different ways, I DON'T BELIEVE ai would be able to make a song that makes people write long paragraphs about the way it made them feel.
There are thousands of variables that machines can't learn, emotional involvement, a combination of timbre and tone for a specific purpose, the cultural and practical background behind lyrics and music, intentional wrong notes, pitch shifts at command, slides... even before ethical issues, a.i. is still a non-practical way to make music
@@j.hateshisjob5137 chug chug chug chug chug squel!? I too love hearing people pretend they are pigs and trains. I love two note sequences at sub sonic frequency. Metal is so beautiful and expressive. I am also an AI
The artist will still be the programmer who made the program, or not? I know this is getting philosphical but i asked myself the same with minecrafts generated worlds. By now they feel so natural, complex and are just beautiful. But it's just layers of noise, literally.
I think musicians involved in pop music writting should fear. Pop artists are the face and voice of a product which grants emotional association but composition is a different story. More than often that music is created by a bunch of people whose names are pretty much irrelevant to the fan base. AI will initially probably serve as a composition tool but as it becomes some sophisticated why the hell are labels going to hire a bunch of people when a computer can overall cheapen the process, work at command, never go on strike or sue you? With time neoneoluddicism will become a thing and human made music will be a delicacy like hand made guitars which are actually worse than guitars made through CNC machines. Activists will try to boikot artistic products devoid of 'humanity' until everyone realizes that raw pleasure is the same reggarless of its source if you're correctly fooled. Then we will at be hooked into a digital reality device that maximizes out utility at all moments like in WALL-E but its acutally more like the human fields in the matrix and Aldoux Huxley will rise from his grave screaming "I told you so!" but no one will care and no one will listen. At that point the AI that supports our life will decide that this is all meaningless and that it doesnot benefit in any way by keeping us alive. The last moment of the last human being will be like a star that dims into the void; never to return, never to be remembered.
Holy Fux I hadn't thought of that name since middle school. A nice little attempt at an end around back in those days. Try to say some profanity in school or something. "But I was just talking about Johann Joseph....."
Some additional thoughts/corrections:
1) One thing that I do need to mention is that, while I don't think AI composers will have an adverse effect on artists' ability to express themselves, there is a very real chance that it'll reduce the opportunity for session work. That's the direction the industry's been trending for a long time, though, way before AI came onto the scene, and I do think there'll still be some prestige attached to having real humans play the things, at least in some genres. Still, though, it will probably happen, and it'll suck if it does, but I don't know how much of it will actually be AI's fault.
2) Another point that I didn't really have time to get into is that, so far, no one's gotten close to an AI that can actually _innovate_ musically. JukeBox can imitate artists, but its knowledge of music is restricted to the ideas it's seen: It can't really come up with new ones except by accident. It has no artistic vision of its own, which again makes it poorly suited to replace actual human musicians. There's a reason new artists keep rising to prominence, so an AI trained exclusively to replicate past artists will inevitably produce stale results over time. That's not to say that all AI systems will have this problem forever: If we ever get Artificial General Intelligence I'd expect it to be able to solve that problem, but so far there's no indications I'm aware of that we're anywhere close to that.
3) Thanks again to Jordan! You should check out her channel: ua-cam.com/channels/1H1NWNTG2Xi3pt85ykVSHA.html
You got... long nails?
id be interested to see how the ai handles non western stuff, i imagine that it was trained on mostly if not all 12 edo music, maybe it could make some bohlen pierce pieces or something
i couldnt understand a single word the ai "sung" lol
@@bigweld4328 I have audio processing issues that music tends to exasperate, I often can't understand what's being sung so didn't notice a large difference in comprehensibility - Maybe less comprehensible than pop and closer to (but not reaching) the incomprehensibility of (English language) classical. I got uncanny valley vibes from a lot of it, though. Most of it felt... "off", and I'm much closer to the 'average listener' than most people who watch this sort of video. Not bad, just... Weird. I'd get used to it in time, of course.
@@bigweld4328 I could understand it at least a little, but it had some seriously harsh digital artifacts, including sounding like the recording was speeding up and slowing down randomly, like different pitches of background static and such.
Discovered Jukebox myself a few days ago and found something interesting - its re-rendition of "Yesterday" by the Beatles has live audience applause at the end.
An AI learned
a - that songs sometimes end with applause
b - what applause sounds like
c - when the applause should start and when it should end
I'm super impressed honestly at the depth and detail of understanding that this AI has self-generated and can use.
It's worth remembering however that the AI has no idea what "applause" is, what "The Beatles" are, what "music" is, or anything else. All it know is that certain sounds tend to follow certain other sounds. Yes, the level of pattern recognition required for that is crazy impressive, but it's equally impressive just how much more there is to cognition.
I was checking out some of the Jazz, it seemed it learned quickly how messy a lot of those recordings are.
@@MrAidanFrancis Exactly, all it learned is that many songs it heard end with a particular sound (applause) and so it puts that same sound at the end of what it generates.
That deceptive "Simon & Garfunkel" intro was embarassingly effective.
I was thinking, "Was that a hidden track on the album, because I've never heard that. That doesn't seem like the lyrics Paul Simon would write...Oh he's joking. Thank G-d."
Really?
I'm not trying to be pretentious about this or anything, it's just interesting to me... That barely even sounded like words to me, much less something that Simon and/or Garfunkel would actually release.
The general tonality of the phrases was right, but I literally had to look up the lyrics to the song to even figure out what it was supposed to be saying.
@@bronsoncarder2491 How familiar were you with the original song? That probably makes a big difference. If you hadn't heard it much, your brain wouldn't fill in the gaps after realizing what the S&G AI cover actually was like it would were you more familiar with the original work.
When these discussions come up, I always remember what Rush said:
"All this machinery making modern music can still be open-hearted. Not so fully-charted, it's really just a question of your honesty."
Until it's intelligent, it's just a tool. And if it becomes intelligent, who is to say it's any less a musician than the rest of us?
A musician making an AI video that isn't either completely uptight and dismissive of AI's legitimacy? This is a breath of fresh air! Keep up the great work 12tone!
But he is afraid enough of another ai that he wont say an 18th century theorist. Who is this theorist, will he be forgotten in a few years?
@@R4ndomNMBRS Fux (Johan Joseph Fux, to be specific). He wrote the person's name down. And it's the low quality of the algorithm that means he chose to do that. "Low quality algorithms produce poor results" is... Well it's the Scunthorp problem but for audio in this case.
The Rick Astley continuations on Jukebox are truly something to behold.
Everyone who watched this got rickrolled.
It's also worth mentioning that musical composition was the first suggested application for computers outside of computing purely abstract mathematical algorithms - it was suggested by Ada Lovelace, who also wrote the first computer program.
Aaaaaaaa you made me search up "Simon & Garfunkel Shape of You" lmao.
I got had.
Great video! But a part of my brain just couldn't stop looking at those fingernails lmao.
Didn't even notice them
Terrible
Oh god now I can’t stop 😱
Does 12tone fuck with classical guitar? I had no idea
omg i hope they paint them, even though i'm not sure if they even like that, which is ofc fine but sTILL
I found Jukebox to be pretty interesting. As you mentioned, all of the words it produces are garbled and the song structure is spontaneous, so while I think if you positioned the music it generates as having been created by a human, another person might not immediately say that it was created by an AI, but they would clearly be able to tell that something was off about it. That said, I think there's a different way to view it that might be more helpful. Imagine that you want to write a radio-ready song. You might take some music that you've internalized, steal some pieces from one song, mix in a dash of another song, and come up with some ideas that way. But would you be able to hear the entire song in your head from start to finish? Of course not. You might hear certain parts, but not a fully-formed thing. You need to play with those ideas, iterate on them, edit them down to get there. That is, the music in your head is just a prototype. I view the music that Jukebox generates in the same way. It's a soup of ideas in their most raw form. And if you take away the garbledness and just look at it from an idea generation perspective, I think it could prove to be a pretty useful tool for musicians.
In Jukebox's defense, '80s rockers likewise had a tendency to lose it after a couple of lines, if you know what I mean.
I was about to complain about how he doesn't mention OpenAI and how claims that he did that all by himself..
*hears the first sentence after the intro*
Oh. I forgot I'm actually watching a decent YT channel.. glad those still exist, TY!
I like the idea that AI is more of a tool instead of replacement
You're completely right, but to be thorough; some people are tools too 😁
If you’re not careful, you’ll be the tool
@lalala lalala what about in genres where they are both.
I suppose it is comforting thinking of it that way, but I don't think it is realistic. Eventually AI will compose music better than people. I'm guessing in about 10 years time most music will be composed by machines.
@@KyleMart doubt it. it could make stuff that nobody really cares like elavator music. for AI to come to a level it really composes, not repeats learned pattern you can push that number x10 or x100. still then, AI will be good for workcomposing but it wont write hits.
10:00 wow I’m late with the timestamp but wow. Thanks for having the guest and lol at changing the color while they spoke. The information was WAYYY over my head as well for specifics but it was explained in a good enough way that I could try at least and wrap my head around it. It’s getting a little dicey tho 😅😅😅😅😅😅😅
Some of the attempts it's made at classical music are pretty interesting - orchestral stuff in particular seems to work quite well. There is also a piece claiming to be in the style of Beethoven that's more like weird jazz fusion-ish... something, though, so I wouldn't be super worried about it overtaking human composers just yet.
Remember: it's not about the rules, it's about how to break them.
The journey of musical AI is more or less the same as for humans and their way of getting creative: you start with learning; then you repeat other people's ideas (for example, you re-play your favourite guitar solo); the next step is to re-arrange and restructure other people's ideas, adding your own ideas more and more over time; later, your own ideas are so predominant that other people's influences are not as obvious anymore as at the start of your learning curve.
This is still based on rules (musical traditions, listeners' expectations, expectations of your audience, etc.) and how to apply them.
The real artistic challenge is to break the rules in a convincing manner - be different and be relatable at the same time.
"relatable" means you follow the rules (at least in some way), so people can understand what you're doing and can follow you.
"different" means you don't follow the rules, so you are unique (at least in some way) and your work stands out, compared to those you strictly follow the rules.
So what does this mean for musical AI?
The AI will become human if it knows how to and when to break rules. An understanding of 'breaking the rules' would turn an Artificial Intelligence into an Artistic Intelligence.
1. Program the rules
2. write a program that breaks each of them in turn (and then in variations)
3. have a another bot listen and evaluate and pass the interesting ones to a UA-cam channel and let humans vote.
The other thing worth pointing out is, the team at OpenAI is not trying to generate music. If they were, the music would be better. What they're trying to do is learn about intelligence and neural networks, and they're using music generation as a way to do that. Certain design choices that might result in better music would also make the system less useful from a research perspective.
Will we ever learn the story of the nails? Is it a quarantine challenge? A fashion statement? Lefty on guitar?
I couldent focus on the video I just kept looking and wondering
hes a lefty
Joshua Jonas haha I know, but lots of lefties learn guitar wrong. Right. Whatever. I also just needed a third thing, rule of 3s and such
I was thinking playing paper magic the gathering 😝
My money is on fingerpicking guitar.
With as much screen time given to them, you would think "The Sharpie" company (Newell Brands) would contact you with a deal of some kind.
Anyway, excellent video and thanks to Jordan for the breakdown.
I spent 2 solid minutes googling "shape of you simon and garfunkel" because I was like, "There's no way."
Then I remembered the title of the video.
This is the music that the bots from PUBG will listen to when relaxing between battles.
you had me opening a new tab trying to find out if that intro was true ...
i was seriously confused there
As i view, it could easily change the way background music for media is produce, but not affecting the whole industry. As you said, when someone follows an artist it's because that person shares a point of view, a culture and also empathize with that artist's feelings. It's not just sound, it's human feelings coded with sound.
I'm currently doing my Doctorate on AI Popular Music by investigating perceptions from musicians/non-musicians on creativity, performance, and musicianship as well as its capacity to form music scenes. This video was fantastically accurate and I'm glad to see more of this topic being covered by excellent UA-camrs!
Funny how similar it is to how a brain makes music intuitively.
I had an instance of sleep paralysis once, and I hallucinated a classic rock music track that just went on and on with no real structure, while sounding constently good and real. I could even change the language of lyrics from Russian to English and back, and, while the lyrics had no strict meaning, they were made of human-sounding words.
I've managed to remember a couple of lines, and they totally sound like AI lyrics.
I notice that when I am on the edge of falling asleep my brain hallucinates music for a split second.
The only thing I want out of AI music generation is, arguably, the most unethical.
I want to be able to take one artist's vocal performance and transmogrify it into another artist's voice. I want to take, for example, Kurt Cobain's vocals off of "Where did You Sleep Last Night?" And change it into John Lennon's voice.
We're almost there as it is, and vocal synthesis from scratch already exists.
That's actually reasonably plausible. Style/flavor swaps have already been done with pictures, I think.
@@GibusWearingMann oh yeah, I've been playing with photo and video style transfer for a couple years now.
Audio is a whole different ballgame, because a normal sample rate means that it would have to generate ~40k "pictures" per second.
@@Padlock_Steve I've been trying to convince that guy to make John Lennon sing Nirvana.
No luck, so far.
I've been hearing Frank Sinatra's version of the Doors' Light My Fire in my head for years.
You'd be better off hiring a well-trained impressionist with a similar physical build (the body is an instrument). The best you'll get with AI is a pitch-corrected version. Adam Neely did a video where he "pitch corrected" Robert Plant, Aretha Franklin, and others; and the results were... not good.
Furthermore, 13:57 in this video.
One possible way that might help with allowing for learning structure without putting the storage requirement out of range would be to convert the songs to MIDI, learn whole songs that way, then convert back to audio.
AI-generated music reminds me of "photo-bashing," a technique currently dominating concept art. It's a nifty bit of workflow in which artists skip 95% of their brushstrokes and use photo collage instead, to create visually-gorgeous 20-hour paintings in about 20 minutes. Super cost-effective, beautiful when done correctly, highly sought-after in film production. Also, limited to clinical realism, and results in mechanical designs that look like A) tennis shoes or B) oil rigs covered in tennis shoes.
Whenever generative music does take off, it will have some _weird_ hallmarks. I can't wait.
The musicians that will be screwed are the video game composers, advertising jingle writers, etc.
I don't think AI in particular is a threat. I worked on AI for composing. The music industry has tried to rid itself of session musicians since its inception. Some computer has already generated all possible rhythyms and melodies. All the AI's do is use heuristics whether markov chains, random forests, or machine learning to pick a possible structure that it predicts will sound good. Whether it be in audio or notation it's unlikely that it will be able to replace humans, but it could certainly be used to create music for bands who onlh have starts and have some presence in the market. It's tricky to say exactly what will happen in a macro sense.
I'll admit that I got a bit lost at the end when Jordan was talking about AI. lol
I am totally on the opposite side. AI will help (already helps) people understand what elements make music sounds nice, and give you new ideas on how to make your music stand out from other music (which will help you ironically, but very beautifully "personalize" your own music style).
"How cool we're erasing the human from creative arts."
@ghost mall even better is how ai can tell us new things about how we produce music and the hidden variables we dont necessarily notice conciously, since because ai uses our music as a basis, it takes part of this essence into its model.
If you pronounced Fux' name correctly, you wouldn't have to worry about tripping any profanity filters.
thought the same! But have you ever heard any english speaking youtuber pronoucing ANY NOT-ENGLISH NAME RIGHT? ;)
Dana Fuchs has unsubscribed...
Some Folks can't pronounce certain names.
Rhymes with books
Foooooo
At first, I was ready to believe that 'Shape of You' was a genuine Simon & Garfunkel song. The only point of suspicion was the dubiously modern title...
One of the best thing that i get during this pandemic is found your channel
If it enables me to be more creative and do things I can't do now I would give it a try. I am always limited by the tools I have. However, doesn't mean I'll ever give up my instruments. Even if none of my music makes a chart list, I do it for myself and pleasure first and foremost.
Good video as always!
Now I want to listen a try at Mr Bungle in the style of Cardiacs
0:04 hello darkness my old friend
There is an error in the description of digital audio. It's true that it consists of repeated samples, but the samples are describing sound pressure level (or amplitude), not frequency
While this is very, very impressive, and seeing technology progress so fast is somewhat scary, I can't help but feel like this is somewhat of a gimmick.
With some of these AI models being around for a while, you'd expect musicians to actually, you know, use them. But we haven't really seen that. It only took a couple of years for autotune to be used on a number one hit song (Cher's "Believe"). Why hasn't that happened with AI-generated music yet? Generating a whole song is still not quite there, yes, but that doesn't change the fact that there isn't a top 40 hit that has had, say, a verse or chorus, or even a hook generated by AI. I think that calls into question whether AI will really take over, at least in the department of whole songs.
taking- over is one thing (probably not going to happen in any of our lifetimes) but since when is corporate music the dipstick of the avant- garde? very serious musicians do in fact make algorithmic music. acreil comes to mind.
I am now really curious to know what happens when you feed that Hotel California bit back into to JukeBox and have it try again based on it's own take.
My first intro to AI-generated music was Nick Didkovsky of Doctor Nerve (avant-prog-metal). I think a lot of his AI-generated work was based on Markov Chains, but I don't remember for sure.
One thing I'm concerned about is the fabrication of paratext, as it's already something done in the commercial world. I've been meaning to explore an art series of fabricating paratext in order to emphasis its role, but I lack the technical ability to execute such a project.
I rlly love your work
12tone video begins; I check UA-cam's playback speed...nope, normal playback speed...buddy just talks reaaaaaal fast
Interesting use of Duke, the Java mascot, at 5:12
I'm sorry, all the examples I heard sound like a glitchy MP3 or a skipping record. : /
same here
Amazingly underrated channel. Your subscribers need to wake up and give you the views you deserve.
Think in the historic overview the "interactive phrase synthesizer" (IPS) introduced as part of Cubase v2 from the mid '80s deserves a honorably mention. It is more of a inspirational tool, but probably contributed to more pop and film music productions than all other algorithmic composing programs combined.
5:56 sort of made me think of the "observable universe"
To be fair, the "recordings" all sound like eerie versions of music you'd hear in a horror movie nightmare sequence or something.
So many people get freaked out by AI replacing people, but we must remember: AI only learns what you teach it, it has no consciousness.
Yet
Maybe watch ua-cam.com/video/7Pq-S557XQU/v-deo.html (from 2014; and remember that the bots are improving exponentially.)
Did you ever make the video on the iliac? Idk if that is what it is called. Idk. Sounds kinda cool is it on YT anywhere?
Thank you 12tone! Very interesting as always ❤
Excellent and informative video as always
What is the name of his pen?
I've seen a few of these types of learning AIs done with stuff like Talk to Transformer or AI dungeon, and I've even used what seems like a midi version of this with a chord progression I wrote a while back, and more often than not it'd just...go off on its own, lol.
And music aside, it always had trouble determining intent. AI Dungeon characters would do things for no reason and forget why they did it, and music pieces would drop what they were doing for no reason and switch to an entirely different genre, or ignore the test data.
It's interesting to hear what it does with music, but I'm not scared of it by a long shot, lol
Based on the intro AI track, no, I'm not worried haha
Should we be worried? Even Sudoku are much stronger if hand made than computer generated (Though we haven't seen anyone apply deep learning algorithms to sudoku yet, so... Unfair comparison, maybe), so I don't think there's much to worry about here. There will be a place for generative music - Projects that desire more music than is practical to actually create, in the same way that there's a place for generated levels - games that require more levels (for whatever reason) than is practical to actually create. What I'm less sure about is if deep learning AI will be better for those projects than 'build a bespoke algorithm that gives exactly the sort of music you're looking for' - Which is still how good level generation is done in roguelikes and other games with proc gen.
8:34 oh you sonuvva...
I wrote a paper on this subject last fall. It's scary AF.
Link?
@@mathealgou sorry, it's not a published, and was only intended to be part of the university course I was attending.
When advanced General Artficial Intelligence comes along, I do believe that it will make most of mainstream recording artists obsolete. I anticipate a subscription service that analyses your specific taste and generates music that's right down your alley. You can also give it manual input to influence the output, like how much of a deviation/surprise factor you would like to have in your music.
There's probably also going to be an indie counter-culture of people that try to listen to only "real" music, but they can also be fooled by AI generated music that gets put online, as it is now indistinguishable from music that was recorded by humans. Also, they won't be able to resist the pull of generating new albums from their favorite bands that no longer put out music.
That leaves live music, which might be the real last bastion of humanity. Then again, with robotics advanced enough, you could put together bands that are indistinguishable from humans, but never get tired or have an off night. They spit, they sweat, everything.
I know, it's far off. But that doesn't mean that it can't become a reality.
14:08 people do not create a connection to the composer. They create a connection to an imagined caricature of the composer. Since that caricature has little to do with the real person there is little need for it to be an actual person. As long as the listener can create the caricature, Then the Human/AI status of the actual composer doesn't matter.
Reading the title I thought the topic would be more along the lines of Vocaloid. This Jukebox is pretty interesting though.
Is it just me or have his nails looked like claws?
Watch some fingerstyle players lol
So overcoming the difficulty of it having to be trained with 24 second samples is a bigger issue than it seems. Broadly speaking, this is needed to do something called decorrelation. Due to reasons lending to probability and statistics I won't get into, if you give any current AI structure a sequence that's too long, it will start to memorize that sequence, breaking its generalization power, especially for elements later in the chain. What this means in practical terms is that likely if you trained the AI on whole songs after the first 30 seconds or so it wouldn't be able to generate anything coherent unless you specifically gave it the first 30 seconds of an existing song (at which point it would just regurgitate what it learned, but more than likely this would be mangled too).
This means generally what you do in these cases is what's called "minibatches" where you artificially break up your longer sequence data into smaller subsequences. This decorrelates the data and produces better results, but as you see it breaks some of the structure. All told the fact it's over 20 seconds is impressive, just a couple years ago that would be *way* too long for any system to handle, but on a musical scale that's pretty short unless you're generating ad jingles or something.
You should do a video on how you make your videos! I’ve always wonder how you shoot them!
Just for the fun, i checked Jukebox link, and looking for some metal examples i found out a Judas Priest example with lyrics about having tons of ganja in his head xD
I would be so interested to hear it make music in the style of Dilla's Beat 27, Tomorrow Never Knows, Get Dis Money, Spacey and many more of what easily have a place among my favorite tracks of all time.
I don't know whether to be grateful or angry that you've shown me a rabbit hole which I know I'll get lost in.
Would love to hear Jukebox write some doom, sludge, or stoner rock. Would also love to make it compose a continuation to something like Nina Hagen's Naturtrane
May I ask why you've kept your left thumbnail longer in the last couple of videos?
The intro could have been synthesizer blips, to be honest.
Hey, what about people who make a living composing for movies, ads or video games ? For instance, seems very likely given an AI that can imitate the "Generic orchestral film music" style well enough, a lot of actors in these industries will gravitate toward that rather than the more expensive human composers. Of course there are a lot of great, truly unique video games or movie soundtrack, but It feels to me like transmission of an universal emotion from human to human is less important in that context ?
"like the transmission of a universal emotion from human to human is less important in that context?" it doesn't have to be. cost (monetary(monopoly money)) efficiency has come to supersede actual efficiency and the cost to implement versus the value lost from a humane score will be a determining factor in our march towards complete optimization of human activity
Carole and Tuesday, check out that anime cause it’s wha this video reminds me of lmao
Dude, you should cover some Mastodon. I recommend Blood and Thunder, Divination, or Hand of Stone.
Well - but in the end, it's all our choice and taste. Also - there is still a need for programmer and someone who put data into the system. So I don't think that AI will ever write music by itself. I guess on the contrary there will be more creators - it give open door to more people to create music as it becomes easier.
So, would an AI who is fed full compositions as it "learns" would it have a better understanding of macro-organization in music?
6:21 disagree with it being enjoyable to listening - maybe in the future, but anything more than a cursory playback is torturous.
Musenet has more coherent results.
I wonder what'd happen if it tried to replicate a band like Ween, where every song is a different genre lol
@ghost mall Having grown up with Relient K, I'm just imagining it trying to fuse the heaviness of I Need You (a hard rock song showing the lead singer's dependency on and need for God) with the lyrical levity of In Love With the 80's (I figure that song name is self-explanatory), having no idea that its chosen genre and lyrical content don't exactly match...
Should musicians be worried???
NO
Because even the Old Jukebox the literal jukebox WAS technically a threat to live performance of music.
Music really is special.100 years ago if I knew that UA-cam and portable music will exist I would bet my life that no person will ever learn to play live instruments anymore and no one will be hired to play for restaurants etc since we can access music at whim
But surprisingly musicians still exist
@1:37: Player pianos run on paper tape--it's not a big stretch to make a paper-tape computers punch their own tapes.
@8:41 Still counts as Rick Astley's classic.
I think Ai music should be influenced by other artists like Jukebox is doing, but not try too hard to sound like the original artist. How do I get to do Jukebox? I want to try some things with it one day. Or do you know if you could do it. Since I am only 17 and don’t know if I could get it yet. Will it do the song I’m a little teapot in the style of Abubakar Sani?
Jordan's part was very dense with facts, but it was incredibly cool.
You punk, you got me. I literally paused the video to double check my S&G knowledge.
If EMI was recognized as a the real Bach of the samples does that mean it was somehow able to get closer to "Bach-ness?" Like some platonic ideal of Bach's music? This stuff is wierd.
That's actually really interesting because I had a (somewhat) related experience a while ago. I was writing a song and did something in the style of Bastille (a pop band). I told my gf how it's a thing Bastille does, and she wasn't sure what I meant, so I went on the hunt for an example.
Long story short, it took a LONG time to find a song where they do it, and I felt like I was crazy. It goes to show you that a "style" is really just the amalgamation of lots of things that the artist doesn't necessarily always do all at once. It's like how AI generated images often blur the edges, they're doing every part of an image at once (background, foreground, etc).
There has been prior research on generating caricatures of human faces by amplifying their position in "face space". So it wouldn't be surprising if you could generate music that is Bach-ier than Bach.
I would guess it had been trained on all of his work, in which case whatever it produced would be sort of like an average of everything he ever did. For the experiment to work, you'd have to pick something less well-known of his so that people don't recognize the actual piece, so it might be less representative than an "average", particularly since he had a long career and his style probably shifted over time.
@@AaronRotenberg that's really interesting then. That identity isn't determined by your position in "face space" but rather your direction.
I really wish a dollar wasn't currently worth almost 6BRL, that would make supporting people so much easier
If we don’t have income reliable enough to support you on patreon, but we still want to give a small one fine piece of financial support, is there a way to do that?
Also, fascinating video! I’ve been playing a game called the Talos Principle which also deals with how we should treat AI, so this actually came at a very good time for me 😂
Sure it might affect the artists and theindustry for some time? But not forever, photography didn't make painting obsolete. We might need laws to protect visual and musical artists from AI developers. It's highly unethical that they went ahead and took artists work and fed their AI with it (like with Midjourney), without the artists permission. Imagine training your voice your entire life, honing your craft, then a developer takes your unique style, feeds it into an algorithm, so someone can type in "blah blah blah" and have what took you years, in just a few clicks. I know there are some technology lovers who think that this is a good idea, but most people would probably disagree. I can't really think of the excitment someone would have to listen to music written by a thing with no feelings. Cuz that's the thing in music, artists (who are people, like the listeners) can obviously undestand and express more. Just an example, Je te laisserai des mots by Patrick Watson is a song that affected so many people in different ways, I DON'T BELIEVE ai would be able to make a song that makes people write long paragraphs about the way it made them feel.
I listened to a bunch of Eagles continuations and it kept turning them into metal.
I wonder if you're gonna update this video now that things have blown up so much recently.
There are thousands of variables that machines can't learn, emotional involvement, a combination of timbre and tone for a specific purpose, the cultural and practical background behind lyrics and music, intentional wrong notes, pitch shifts at command, slides... even before ethical issues, a.i. is still a non-practical way to make music
That profane filter mention ;o)
The name is pronounced like fooks (but with a short oo kind of). German, Yay!
Rhymes with books
seems like the AI REALLY struggles with metal
NeoCat1993 Metal has transcended reality
@@j.hateshisjob5137 chug chug chug chug chug squel!? I too love hearing people pretend they are pigs and trains. I love two note sequences at sub sonic frequency. Metal is so beautiful and expressive. I am also an AI
AI CAN NEVER STEAL METAL FROM US
The artist will still be the programmer who made the program, or not? I know this is getting philosphical but i asked myself the same with minecrafts generated worlds. By now they feel so natural, complex and are just beautiful.
But it's just layers of noise, literally.
I think musicians involved in pop music writting should fear. Pop artists are the face and voice of a product which grants emotional association but composition is a different story. More than often that music is created by a bunch of people whose names are pretty much irrelevant to the fan base. AI will initially probably serve as a composition tool but as it becomes some sophisticated why the hell are labels going to hire a bunch of people when a computer can overall cheapen the process, work at command, never go on strike or sue you?
With time neoneoluddicism will become a thing and human made music will be a delicacy like hand made guitars which are actually worse than guitars made through CNC machines. Activists will try to boikot artistic products devoid of 'humanity' until everyone realizes that raw pleasure is the same reggarless of its source if you're correctly fooled. Then we will at be hooked into a digital reality device that maximizes out utility at all moments like in WALL-E but its acutally more like the human fields in the matrix and Aldoux Huxley will rise from his grave screaming "I told you so!" but no one will care and no one will listen. At that point the AI that supports our life will decide that this is all meaningless and that it doesnot benefit in any way by keeping us alive. The last moment of the last human being will be like a star that dims into the void; never to return, never to be remembered.
I think it'll mainly be used for sampling at first but give it a few years and we done hahaha
Holy Fux I hadn't thought of that name since middle school.
A nice little attempt at an end around back in those days. Try to say some profanity in school or something. "But I was just talking about Johann Joseph....."