AI and music

[darkoshi]

In media of the past, artificial intelligence was usually depicted as being more intelligent than humans. Being able to do complex calculations rapidly; having vast stores of data for use in those calculations; being able to come up with logical and correct answers.

The problem with current LLMs is that they are too human-like in their intelligence. They are only as smart as the data they've been trained on. They can't tell you how they know what they think they know. They may make things up without even realizing it.

I am impressed with some of their other abilities though, such as generating images.

..

On YouTube, I saw a recommendation for "Flatulence Flamenco". The video thumbnail showed the word "Falauence", so I wondered if it was a typo in the video name. The song lyrics on the video page proved it not to be a typo. It is almost certainly an AI-generated song along with an AI-generated thumbnail image. The lyrics are amusing to begin with, but quickly become gross. Not something I'd want to listen to.

..

Another time, YouTube gave a recommendation for a remix of a song by an artist I like (Illest of Our Time by Ren). The remix is very good. I may like it better than the original song.

I listened to a few other songs from the channel that posted the remix, but didn't care much for them. The videos on the channel all had AI-generated artwork. At that point, I realized that the channel's remixes including the one I liked so much must be AI-generated too. I wasn't sure how I felt about that.

I saw that the person who posted the remixes responded to many of the comments on that video. I wondered if their comments were AI-generated too. But the comments seem real. The person even mentioned using (AI) models in the making of the remixes.

So how do I feel about that? Not as impressed by the remixer's skill as had they not used AI, but still appreciative of the result they came up with.

Comments

Even better than the real thing

[frith]

In a time in which alternative facts and plausible confabulations proliferate while fact checking gets drowned out or cut to save on expenses, AI probability texts are going to make everything worse. At least the visual art can be pretty and visual art is already a confabulation.

I used to get emails from relatives with forwarded falsehoods. I'd use Snopes to reply with what error they'd been fed. At least the emails stopped.

Re: Even better than the real thing

[darkoshi]

I wish there was an option in my search engines to filter out results linking to sites with a lot of AI-generated text content. Hopefully they'll figure out a way to do that without blocking legitimate sites. I wonder... I remember there being site-ranking add-ons based on users' feedback. Something like that might work. In the past I didn't feel I needed that functionality but maybe it would help.

Re: Even better than the real thing

[frith]

At least on Dreamwidth the owners and their volunteers smite the Search Engine Optimization spam and there are no LLM-generated blogs (what would be the point? there's no revenue to be generated by blogging on Dreamwidth). What is needed would be web-rings of not-for-profit or at least ethical sites and get our word-of-mouth opinions and recommendations from that. Like, a pre-Google, pre-surveillance capitalism environment.