ChatGPT on Classical Music

[Florestan]

Otoh, this:

"Sonata in D major for Two Pianos" by Dmitri Mendeleev - Mendeleev was a chemist and the creator of the periodic table. He also wrote music, including this sonata for two pianos, which he dedicated to his wife.

looks entirely fabricated.

Mapman, could you please ask ChatGPT to name the best recording of this sonata?

[relm1]

Otoh, this:

"Sonata in D major for Two Pianos" by Dmitri Mendeleev - Mendeleev was a chemist and the creator of the periodic table. He also wrote music, including this sonata for two pianos, which he dedicated to his wife.

looks entirely fabricated.

Mapman, could you please ask ChatGPT to name the best recording of this sonata?

That is quite interesting actually and maybe a clue in to how it works.  Mendeleev was closely connected to Borodin, composer/chemist who was championed by his wife, a pianist.  I wonder if the Sonata in D major for Two pianos is the one by Mozart?  So, it's tangling different topics that have a small connection to create its unique response but here getting it all wrong because it doesn't actually understand how the connections work.

[Florestan]

That is quite interesting actually and maybe a clue in to how it works.  Mendeleev was closely connected to Borodin, composer/chemist who was championed by his wife, a pianist.  I wonder if the Sonata in D major for Two pianos is the one by Mozart?  So, it's tangling different topics that have a small connection to create its unique response but here getting it all wrong because it doesn't actually understand how the connections work.

You might be on to something. Very interesting.

If you ask me, AI is never going to replace human intelligence. Like in , never ever.  It's not even at the level of children, because children are often able to make very interesting points and links and even when they are mistaken they are humanly so --- which is not the case with AI.

[Todd]

Remember that one of the largest corporations in the world is pouring billions into the parent company of ChatGPT.  It is in an early stage and the goal is for the investment to be profitable.  That means expanded contracts in the customer service space (with projected annual growth rates of 25%+ for years to come, though from a small base), the software development space, and, of course, the juiciest plum of all, the military.  There will be many other profitable deployments, too.  Blabber about AI matching human intelligence and the singularity and so forth is either salesmanship or self-delusion.  Let's start with the basics: define human consciousness and intelligence.  Once you can definitively do that, then maybe you can architect real AI.  People have pondered AI in some vague form since at least the 1870s, and we ain't even close to HAL yet, so, there's that.  ChatGPT is the world's most sophisticated plagiarism tool.  World class devs can disabuse me of this notion if they want.

[71 dB]

If you ask me, AI is never going to replace human intelligence. Like in , never ever.  It's not even at the level of children, because children are often able to make very interesting points and links and even when they are mistaken they are humanly so --- which is not the case with AI.

Should people ask you? Are you an expert in artificial intelligence?

If you ask me, it is a question of when rather than if AI surpasses humans in intelligence. The humankind keeps becoming dumber and dumber day by day and the target for AI in that sense becomes closer all the time. Then again, I could be wrong since I am not an expert...

[Florestan]

Should people ask you?

"If you ask me" is a figure of speech. Whether people should really ask me or not is irrelevant.

Are you an expert in artificial intelligence?

Time and again, this effing pernicious expert myth...

Look, if only experts had the right to speak on any given subject, then you'd have no business talking about Mozart's operas at all since you're not an expert on them, okay?

The humankind keeps becoming dumber and dumber day by day

This logically implies that you are becoming yourself dumber and dumber day by day --- unless you claim you don't belong to the humankind.

and the target for AI in that sense becomes closer all the time.

The funniest thing of it all is that AI is being developed and tweaked by the self same humankind which according to you gets dumber and dumber by the day.

Then again, I could be wrong since I am not an expert...

I could compile a long list of things that experts claimed to be true at some point in history and which were subsequently found to be false by other experts at some future point in time --- and vice versa.

Experts are hugely overrated.

[Uhor]

The humankind keeps becoming dumber and dumber day by day

Quite the opposite at least in some privileged places.



If we are speaking of emotional intelligence (EQ), it might be too new to quatify but my personal opinion is that, on average, we are better.

[71 dB]

This logically implies that you are becoming yourself dumber and dumber day by day --- unless you claim you don't belong to the humankind.

No it doesn't, because mankind is not the same thing as an individual (such as you or me). Mankind is a mixture of smart and dumb individuals and everything in between. So, what I said means the mixture of individuals becomes dumber. This can happen in many ways meaning there is no logical implications for what is happening to your or my intelligence without addition information. What if the additional information is that the new generations are dumber than the old ones?

The funniest thing of it all is that AI is being developed and tweaked by the self same humankind which according to you gets dumber and dumber by the day.

Individuals aren't becoming necessarily dumber and to my understanding people who develop AI tend to be smart.

I could compile a long list of things that experts claimed to be true at some point in history and which were subsequently found to be false by other experts at some future point in time --- and vice versa.

Yes, but this is because predicting the future is hard. You said AI will never ever reach the level of human intelligence. That is quite a bold prediction if you ask me. I believe AI will at some point surpass human intelligence, but I don't know when. That's the big uncertainty part. In 30 years? In 50 years? In 100 years? In 200 Years?

Experts are hugely overrated.

No they are not if they are real experts and not hacks.

[71 dB]

Quite the opposite at least in some privileged places.



If we are speaking of emotional intelligence (EQ), it might be too new to quatify but my personal opinion is that, on average, we are better.

Why are those curves straight lines? How about the years 2005 - present?

[relm1]

Why are those curves straight lines? How about the years 2005 - present?

The data on the graph says 2007.  I think we got stupider after that.  I'd interpret the straight lines as limited data points.  For example, this probably isn't a yearly data sample but maybe two data points hence a straight line.  It's a bit of a bogus chart.  I guess they anticipated those reading it have become stupid.

[Uhor]

This rise in intelligence is called the Flynn effect, just in case you want to investigate in Google Scholar for meta-analysis or whatnot.

Here's another graph I found.

[Mapman]

Here's an interesting video about attempting to use ChatGPT for composition. (It isn't particularly helpful.)

[Spotted Horses]

Mostly I notice people claiming ChatGPT has magical powers, or is crap. I've used it recently and it seems it is mostly like using google, except the answer is organized into preachy paragraphs instead of a list of hits.  It is remarkable that it can more or less parse ordinary English, find stuff in it's enormous library of text and extract out and format something relevant.

I decided that listening to a set of works in the order composed gets tedious, either listing from beginning to end, or vice versa. So I used ChatGPT to make a random playlist of Haydn Sonatas. I asked it to put the numbers from 1 to 62, excluding 21 to 27, in random order:

Here are the numbers from 1 to 62, excluding 21 to 27, in random order:

10, 37, 42, 1, 60, 45, 11, 5, 55, 48, 36, 13, 62, 56, 35, 38, 33, 2, 52, 9, 18, 12,
44, 20, 3, 58, 54, 17, 51, 19, 31, 32, 6, 59, 14, 30, 46, 4, 47, 50, 7, 34, 49, 8,
40, 57, 41, 61, 53, 22, 25, 28, 39, 43.

These numbers are in random order, excluding the range from 21 to 27.

Here's my Haydn Sonata playlist.

[DavidW]

That is kind of overkill to use AI to simply generate a random permutation.

A colleague of mine has a paid subscription to use gpt-4 and he claims it is much more sophisticated than the free gpt.

[Spotted Horses]

That is kind of overkill to use AI to simply generate a random permutation.

How would you do it? Saved me screwing around in various programming languages.

[Florestan]

How would you do it?

The old-fashioned way: write the numbers on pieces of paper, mixed them in a bowl, pick one by one randomly. 

[DavidW]

How would you do it? Saved me screwing around in various programming languages.

There is a random permutation generator that I use online.  I only use it when I'm scrambling multiple choice questions when I'm writing a test.  Here is one.

[Spotted Horses]

The old-fashioned way: write the numbers on pieces of paper, mixed them in a bowl, pick one by one randomly. 

ChatGPT took me 5 seconds, a lot faster than preparing the slips of paper. But then I had to cut and paste the output to a text file and import into a spreadsheet, which took me a few minutes.

There is a random permutation generator that I use online.  I only use it when I'm scrambling multiple choice questions when I'm writing a test.  Here is one.

But that does a contiguous range a numbers, it doesn't allow you to specify a gap. I'd have to remove the missing sonatas by hand.

If I didn't think of ChatGPT, I'd have written a program. Generate the list of 55 sonatas, generate a random number between 1 and 55, pick the number out of the list. Generate a random number between 1 and 54, pick the number out of the list, iterate until no numbers left. That would have been annoying.

[71 dB]

I decided that listening to a set of works in the order composed gets tedious, either listing from beginning to end, or vice versa. So I used ChatGPT to make a random playlist of Haydn Sonatas. I asked it to put the numbers from 1 to 62, excluding 21 to 27, in random order:

Wait a minute...

You have to ask ChatGPT to give you a list where the works actually are in the order Haydn wrote them!

[Spotted Horses]

Wait a minute...

You have to ask ChatGPT to give you a list where the works actually are in the order Haydn wrote them!

Read again, I asked ChatGPT to give a random list of Sonatas 1-65, excluding 21-27.

My experience that listing to a long series of works in the order composed can be tedious because works composed at the same time don't have stylistic diversity.