Of course the strict adherence to soundchip limitations makes chiptunes sort-of immune to AI Music generators for now, but I'm also talking about music in general.

I feel like the amount of AI generated Music has significantly increased recently. In the past I had no issue leaving a comment calling people out for it. I'm starting to doubt my ability in telling apart AI generated Music from real Music however. And I would be forever ashamed for calling a real song AI slop.

(there is also the area in-between: Music made partially by AI and partially by a Person. I don't want to discourage people getting into music, but them using AI will inevitably stop them from improving imo.)

As far as I can tell many music generators are still a form of diffusion, just in a series of auto-regressive blocks, and therefore have a high noise component / low transparency with instruments morphing into each other, sort of like early image generators blending details together. Apart from that there are certain structures similar to the weird sentence choices of Large Language Models. But I just can't identify either of those signs clearly enough anymore to write the songs off as slop.

Honestly, you've got all the hallmarks mentioned here already. Noisy vocals (and drums especially!), instruments blending together, stilted lyrics, and basic structure are pretty much all of them.

A few weeks ago I busted a dude in a random discord for an obviously AI generated song

. It was super funny too, the dude was doubling down in VC claiming to have written this while conceding that:
- He didn't remember any plugins used to make it
- This was his first and last song
- "The guitars were made with FL"
- His mic in vc sucked

I've found that context in light of the previous factors really is the biggest tell for fully AI-generated music. If it's out of nowhere from someone who wouldn't have any right making something that sounds like that, it's probably AI. Massive production or compositional "skills" out of nowhere, especially if it appears to be very studio-polished on first glance, an inability to answer basic questions about the compositional and mixing processes, etc. If you think someone's faking it, press them with questions only a legit songwriter could answer.

I use BlockTube to block entire AI channels and haven't seen any in a while.

Any channels that are called like "Mario Citypop" and with a suspicious thumbnail.

Also, I don't care what it sounds like. I don't want to analyze what it sounds like. I wish people would focus more on discussing human-made things.

I don't want to analyze what it sounds like either. I know that every bit of attention I give to AI is attention I should've given to a person instead.

The problem is that not every bad actor is so blatant when it comes to making AI content. I fear that If I ignore the issue for too long it just disappears because I can't tell Reality from AI generated stuff anymore.

I honestly feel like trying to identify what is or is not AI-generated is an uphill battle that can't really be won. it's hard not to make it a witch hunt. Tex and Matt are both right that if you're in a place where it could crop up, context matters a lot. if something seems suspiciously inhuman or devoid of humanlike context, then the best you can do is ignore it. the truth of the matter doesn't really come into play there. you had a gut feeling, you acted upon it, and there is no shortage of other stuff to think about trying next. but as tech grows and trends move forward there are always going to be new and different "tells", and old "tells" are going to fall out of favor.

I guess for me personally I don't go looking for music in any place that this would ever be a problem. my discoverability paths usually involve branching around from artists/bands related to artists I know; or labels or genres I'm reading about and exploring via sites like RYM, Wikipedia or Discogs. I've been doing it this way since long, long before the "AI" boom, so I don't claim that this is an intentional method of avoidance on my part. in terms of finding new music though I don't think AI-generated music affects me in the slightest. my methods are pretty human-centric.

for music it is extremely easy to prove to someone that you are not in fact slopping it up just by showing a project file. so you can just shoot them a message and if they are all hesitant and defensive and stuff then they're probably a big stinky liar. google has SynthID for watermarking their ai's outputs, we need HumanID. if you're drawing, take a timelapse! even just showing that you made SOMETHING with a pencil in the past is so much better than a whole gallery of perfectly-rendered art with no visible brushstrokes.
ok yeah that was kind of a tangent. this is about music. i just mean to say in my very awesome and cool personal opinion I think that it is okay to be a skeptic and if you are making art with your own two paws, you should make that apparent. no wait i meant to say breaking the pencil is 67 TUFF save the trees and be mango blud not diddy foid. epstein israel respect WE CAAAARRY THE FLAME
but yeah context is absolutely the best indicator here. i don't hate ai nearly as much as i hate big fat stinky poopoo liars sorry if i sound like a paranoid schizophrenic

simple: dont listen to music

The thing to not use is any solution that claims to "identify" AI material, chances are these are also powered by AI and will most likely produce a wrong result, leading you to be misled.

A lot of what is mentioned in this thread is true, however, but there's one very key tell-tale sign for me: Upload frequency.

A real artist would upload, at most, a set of songs at once all belonging to one album. However the time between these could be months, or for single songs could be days or weeks. They would most likely also be on differing subject matters, and not the same subject, right?

AI music channels, especially those "cover/remix" channels, they'll upload dozens of remixes in a day, then upload another dozen the next day. These aren't even ran by humans, but rather by content farms. It is simply just a new wave of "Pregnant Spider-Man and Elsa" I think.

As for songs themselves, keep an ear open, you'll hear some... weird choices, like scratchy or grainy sounding vocals, not intentionally but as a result of some artifact, instruments morphing like was mentioned in the thread earlier, and particularly for remixes, completely changing the direction like a minute in. It's also typically going to try sounding incredibly well produced, with "top-tier vocals", "punchy big drums", and so on so forth.

Others in the thread are right, though, the tells that exist now may just not exist in the future, and some even mitigated now by smarter actors. The thing is, smarter actors typically don't really engage with this stuff, I feel. And, I'd probably argue that tells may become easier in the future, as AI algorithms feeding themselves with AI-generated content will slowly poison them.

But maybe all of this isn't even needed.

The biggest thing you can do is surround yourself with known human artists - people on this site offer as great examples, and who they follow as well more than likely too. We're also not shy from providing imperfect art, as we ourselves are not perfect. Unless any of us suddenly decide to be AI evangelists, it's unlikely, no, nigh impossible you'll see AI generated anything from us either here on BotB or like, on any of the social media websites that we're on.

How do you identify/avoid AI-generated forum posts?

Everyone should just make oscilloscope videos.
Pretty hard to fake oscilloscope stuff with multiple channels.

music made before aí

"made before AI" is hard to pinpoint because we had generative music in the 80s (edit: may have actually been the 90s. will fact check). There was an old DOS experiment that was trained on just Bach's catalog and would generate countermelodies to user input. AI music just hasn't caught on till now

When I find new-to-me things it's nearly always because about them in the context of something else and thought "that sounds like it might be cool." And if I wind up liking it I usually want to know more than that. So I guess who- or whatever is pumping out the slop would have to be really determined to spoof on a non-musical level. Add that to me being a grouchy ass who listens to vanishingly few things made post-y2k and I feel pretty safe for now

also #234164 Certified Um Ackshually Moment

I also listen to very little of anything modern, much of it from places like this lol

The little AI stuff I have heard has always the same issue : absolutely terrible actual sound quality. It tries to faithfully recreate low MP3 bitrate artifacts, variety of analog noises and distortions and things like reverb and other effects have really strange sound to them without much "space" in it. It is especially apparent in headphones, to the point I think it is unlistenable...

@stupe I apologize for sharing the history of auto-generated music

Before I listen to anything, I ask the music to tell me if I am a crosswalk. If they can't answer I know that the music is AI.

Nuclear crisis diverted. You are now registered to use club penguin.

Looking at the spectrogram of a song you suspect to be AI-generated, you should see a lot of unusually straight lines where they shouldn't be, in things such as vocals, instruments with lots of pitch bends (guitar, solo strings, woodwinds/brass, expressive synths etc). My boyfriend has saved at least a few artists from being drowned in false accusations by providing concrete proof there are no perfectly straight lines in their spectrograms, and has exposed people for the opposite.

currently its easy because I have producer brain so im always thinking about how music is produced and stem separating in my head (thanks ytpmv for ear training me to be able to do that) and AI music almost always sounds like there are lots of very strange decisions that would be a really big pain to do for no real benefit (i.e. very slightly inconsistent drum samples, reverb on a synth coming in and out at random, and the biggest one for chiptune especially is the very strange tinny imperfect square wave you can hear in all of the recent sabrepulse stuff lol). but it's been difficult for drumless ambient music for a while which was the main reason I stopped using spotify as it kept recommending stuff from ghost artists which may or may not have been ai but were from people who probably didnt exist (i.e. 0 internet presence outside of streaming). and yeah, tech will either get better or the imperfections will be in different places in the future so there really arent any hard line rules which will always be true

I think the method is actually just to avoid algorithmically recommended music and use human recommendations and tools like RYM and Discogs to manually seek things out. Other than that project files are also a guarantee right now

Where AI excels the most is in doing averages, its hard to contest this fact because of how GenAi works. So imo, just by listening the most weird, unique, thoughtful voices on the medium you are avoiding a lot of slop. The same happens with small niches, if AI is used there, it will be more noticeable than in an american pop hit style song.

This opinion and mindset is not perfect, but I prefer doing this than waste my time identifying AI in average sounding music, and even less because is music that I am not really that interested in support or listen to in the first place. I prefer to protect this community, for example, and here certainly AI is easier to spot and it will be like this until chiptuning becomes the mainstream.