Wolfram Alpha Architecture | Hacker News Reader

Wolfram Alpha Architecture

highscalability.com

13 pointsdrp19 comments

Misleading title again, no really interesting details...

I thought the claim that they've created the 44th most powerful supercomputer to power their searches was rather interesting.

Agreed. No new information here, and zip on the architecture.

His other web site he mentions here:

http://tones.wolfram.com/

is just pathetic.

I wrote a jazz improvization machine 18 years ago which played more sensible music some jazz fans even kind of liked, but it was as useless as this "new kind of" crap.

Music is an abstract language we, humans, speak to express ourselves. You can't touch hearts and souls by generating random sequences within a given scale.

I hope Alpha turns out to be more useful, but good Lord, this man's narcissism case should be included in psychology textbooks.

>Music is an abstract language we, humans, speak to express ourselves. You can't touch hearts and souls by generating random sequences within a given scale.

What if you randomly generated something that was identical to a famous piece of music? Is there a reason why that couldn't happen? And, if not, recognize that it doesn't actually depend on the prior human creation of the piece...so I don't see why, in theory, you couldn't randomly generate (or grow via GA or whatever) a good piece of music.

Music is not just notes. With the exception of some of the modern genres (dance/electronica mostly) music is performed by people and truly good music has nuances not seen in the scores.

There are performances of Bach that I'd turn off after 20 seconds, and there are those I keep in my music collection for years, if not decades. Performaces differ dramatically.

Or take Pink Floyd's Wish You Were Here - was it the scores, or rather Gilmour's unique articulation on the guitar that made us listen to this album? How about John Coltrane's emotional outbursts on the sax? Nina Simone's cosmic-deep vocals?

Music is a far more subtle art, not yet understood from the point of view of Cognitive Science. What random notes are you talking about?

If you keep generating random series of tones, eventually you will end up with Pink Floyd's Wish You Were Here or any other piece ever recorded.

That's probably a very clever way of producing music, but I don't want to wait billions of years until an awesome performance of Mozart's concerto No.17 emerges from your machine. I want it now, please. Thank you.

Right, but with good heuristics it wouldn't take billions of years to generate a good piece of music. Anyway, there's a difference between something being impossible and something requiring a large amount of time.

Heuristics - you mean, you know how we perceive music? Can you explain what makes a piece of music good? Same for performance - why are some performances of the same piece of music better than others?

If I could explain what made a piece of music could I could write a direct algorithm. The kind of heuristic I mean is something that would rule out the vast majority of dissonant crap that has no chance of sounding good.

A lot of what was believed to be "dissonant crap" some 50 years ago is now part of jazz and avantgarde. Many jazz fans value Thelonious Monk's music, for example, more than any popular harmonious crap. Obviously you are not familiar with music. Good luck in exploring it before you try to formalize it.

I think you are still missing the point. Would you not agree that in the entire space of all possible sequences of noise, the vast majority could never be considered music?

Ok, you can omit noise and leave only the chromatic scale (12 notes). In this space you will have to go through all possible combinations of notes in a chord and then all possible sequences of chords for some, say, 16 bars, plus you have to generate the theme on top of your chords (or the theme first and then the chords - that will do too). Then possibly variations, improvizations, and so on. Then there may be deliberate "detuned" elements, like in Gilmour's solo improvizations or Nina Simone's vocals. This looks like a huge tree for your search.

To add complexity, in some modern genres like electronica musicians also express themselves using timbre dynamics combined with notes.

But apart from that, there's a problem with identifying automatically which of your variants you generated are good, which comes down to what is good music.

Then performance. We usually appreciate good performance and a lot of people are not interested in hearing precise robotic play. Again, what makes a performance good? Nobody knows, there's no theory.

Having said that, what's the point of doing tree search at all?

I dont get what walpha has to do with ankos. Is it just for the hype? I doubt walpha uses cellular automata to create the answers.

im amused that the highscalability.com domain being linked to is not responding to my web request. timed out too many times. i'll heed ulf's comment and move on. if it's that good, i'm sure it'll come up 20 more times....

The interesting part of their article is a trivial rehash of this one:

http://blog.wolframalpha.com/2009/05/12/the-computers-poweri...

For co-located servers who pays the energy bill?

Usually the data center charges a certain rate for each circuit and the customer can consume as much or as little as they want (short of blowing the breaker).

Reply on news.ycombinator.com