Jump to content
IGNORED

AI - The artificial intelligence thread


Recommended Posts

  On 7/2/2023 at 9:22 PM, Rubin Farr said:
  Quote

In a talk arguing for the continued importance of “symbolic music generation,” Julian Lenz of AI music company Qosmo pointed out that raw audio models aren’t yet good at grasping the basics of music theory. For example, Google’s MusicLM, a recent general music model trained on hundreds of thousands of audio clips, has trouble distinguishing between major and minor keys. (emphasis added)

Interesting. Would have thought that this was easy (they’re talking about MIDI so the amount of data should be more than manageable).

Then again, major / minor is only a thing in Western classical music, so maybe it’s difficult for human brains too.

  On 7/3/2023 at 8:11 AM, Limo said:

Then again, major / minor is only a thing in Western classical music, so maybe it’s difficult for human brains too.

where tf is @brian trageskin

"difficult for human brains" - we need him to go far, far down this rabbit hole...thankfully these AI thingy's haven't been fed his watmm posts yet - like this one https://forum.watmm.com/topic/102229-harmony/. then we'd be in danger of letting them get their digital minds on all this human brain powered musical knowledge...

  On 6/26/2023 at 8:58 PM, zero said:

the concept of creating an AI deity to help keep all the eventual AI bots in check to I thought was unique. if they are super duper smart tho, they'll probably figure out it was just something made up to keep them under control...and from uh wiping out all the humans. 

  On 6/26/2023 at 9:59 PM, cichlisuite said:

it's like you'd have a bomb in case a bomb goes off.

 

  On 7/3/2023 at 3:39 PM, Summon Dot E X E said:

There's just so much you can represent in the audio space, and the training files are so large, hence training is slower and more expensive

Right, but they said the major / minor thing was tested on MIDI files, which are really small. So whatever the reason is, it's not this.

  On 7/3/2023 at 3:39 PM, Summon Dot E X E said:

I can't help but think that there may be insufficiently descriptive metadata for a lot of music, too, which is key.

You might be on to something here. For image recognition (which ML is very good at) the metadata is clear and descriptive. For language models the training data itself is the metadata. But how would you describe a piece of music accurately? If this was easy to do, music critics would be out of a job.

  On 7/3/2023 at 9:08 PM, Limo said:

But how would you describe a piece of music accurately? 

like this, of course: 

 

GHOST: have you killed Claudius yet
HAMLET: no
GHOST: why
HAMLET: fuck you is why
im going to the cemetery to touch skulls

[planet of dinosaurs - the album [bc] [archive]]

psychology today puts out a lot of trash articles, but this one is good: 

On Superfluous People

  Quote

As if to purposely provoke like a novel kind of adversary, author Yuval Noah Harari has said that "The most important question in 21st-century economics may well be: What should we do with all the superfluous people once we have highly intelligent, non-conscious algorithms that can do almost everything better than humans?" The implication is immediate: if humans aren’t economically viable, they are, ipso facto, superfluous.

 

GHOST: have you killed Claudius yet
HAMLET: no
GHOST: why
HAMLET: fuck you is why
im going to the cemetery to touch skulls

[planet of dinosaurs - the album [bc] [archive]]

  On 7/3/2023 at 6:37 PM, iococoi said:
Expand  

LOL

do you have a tweet or some other social media source i can link to? i wanna share this with a bunch of people ?

  On 7/3/2023 at 10:56 PM, Satans Little Helper said:

LOL

do you have a tweet or some other social media source i can link to? i wanna share this with a bunch of people ?

https://ifunny.co/video/elon-al-is-far-more-dangerous-than-nukes-al-1E4JOcnp9?s=cl

but the video itself is from robotsdraw

  • 2 weeks later...

From Reddit:

ChatGPT got dumber in the last few months - Researchers at Stanford and Cal

"For GPT-4, the percentage of generations that are directly executable dropped from 52.0% in March to 10.0% in June. The drop was also large for GPT-3.5 (from 22.0% to 2.0%)."

https://arxiv.org/pdf/2307.09009.pdf

  On 7/19/2023 at 3:41 PM, Summon Dot E X E said:

Finally, someone else that understands the greatest threat we all face: the Moon.

 

Releases

Sample LIbraries

instagram

Cascade Data 

Mastodon

  Reveal hidden contents

 

  On 7/3/2023 at 9:39 PM, luke viia said:

psychology today puts out a lot of trash articles, but this one is good: 

On Superfluous People

 

So, where in the stage where humans are only seen as a resource (not that hasn't be the case for awhile already) to be exploited to make stonks go up. If humans are not good for that they are a burden. Great, just great.

Rc0dj.gifRc0dj.gifRc0dj.gif

last.fm

the biggest illusion is yourself

Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×