I am pumped to write this piece because for me it’s a nod to cutting edge music distribution facilitated by AI. For now it’s human-centered versus big-company centered. In short, it’s the clever little guy beating the man at its own game. At least for a little while …
Also it’s a lesson about how you can talk with AI to get the answers you want. Sometimes that means stopping after the first response and sometimes it means proving a case to the AI. This post is about the latter.
It’s unlikely you’ve heard of the artist “Tony” unless you are one of 1,007 people in the world (like me, accidentally). Now, I don’t know who Tony is either but on January 18th of this year I added Rolling Stones to my Liked Songs on Spotify.
I didn’t pay any attention, just added a song that I liked.
Until yesterday, when I was vibing out to the song, looked at my phone, saw it was a song by Tony and scratched my head.
It sounded like Don Toliver but it said Tony … so I turned to AI.
Tell me more about this song called "Rolling Stones" by Tony, which is composed by Scott Muscoody, who I think is a known person. Remind me if that's true. It also says the main artist is David, so what kind of project is this? There's a lot of names going on here. Are they using AI to put music out?
A few things are likely going on based on how a lot of new uploads and AI‑driven tracks are labeled on Spotify/SoundCloud/YouTube right now:
The “main artist” vs “featuring/with” names:
Many distributors let you tag a “primary” artist (in your case “David”) and then list other contributors (vocalist Toni, writer/producer Scott) separately; this can make it look like the song belongs to David’s project even if Toni is the main voice.Odd or unfamiliar writer names:
A lot of AI‑assisted or library‑beat tracks credit obscure “composers” whose names don’t appear anywhere else, which is often a sign of either a made‑up pen name, a production‑library alias, or a small, unknown producer.AI music angle:
There is an exploding ecosystem of AI‑aided music where people generate vocals or instrumentals and then push them through cheap distributors under semi‑random artist and writer names. This makes it very common to see tracks with combinations of names that don’t map to any real‑world discography or social footprint.
Rolling Stone by Tony is the name of the song listed on spotify
So you’re probably looking at either:
A small catalog / “mood music” operation, possibly using AI tools for some or all of the production, or
A human‑produced beat/vocal track pushed through low‑touch distribution with messy, semi‑random credits.
I pull out the big guns
in the notes it says composed by Scott Mescudi. That is Kid Cudi!
You’re right that Scott Mescudi is Kid Cudi’s real name, but in this context it’s almost certainly being mis‑used, not indicating a real Kid Cudi composition credit.
Why this is almost certainly not a real Kid Cudi cut
Kid Cudi’s legal name is Scott Ramon Seguro Mescudi, and that’s how he’s credited on major releases.
A genuine Cudi‑written track would normally show up with clear label info, PRO/publishing data, and likely be tied into his known catalog and collaborators, not buried on an anonymous “mood music vol 1” playlist with an unknown Tony/Toni as the performer.
Low‑tier distributors and sketchy uploaders sometimes put famous names (writers, producers, or even fake “feat.” tags) into the metadata to boost perceived legitimacy or game search—using “Scott Mescudi” as “composer” is exactly the kind of thing you’d expect from that behavior.
AI dug in its heels 👠 but I charged on
You seem very sure of yourself?
You’re absolutely right to push on this — I’m confident about the direction, but it’s based on how credits and distribution normally work, not on secret access to the backend of this specific track. Here’s what I can say with high confidence and what’s still uncertainty.
What’s solidly verifiable
Kid Cudi’s real name is Scott Ramon Seguro Mescudi, and he’s credited that way as a writer on major releases.
His songwriting credits are attached to big, well‑documented records (Kanye, Travis, his own albums, etc.), and those works are tracked in discographies and writer‑credit databases.
A tiny “mood music vol 1” compilation with an unknown Tony as artist does not appear anywhere near his documented catalog or known collaborations.
From that, I’m highly confident this track is not part of Kid Cudi’s official body of work in any normal sense.
Is this an "AI song" with synthetic voices and mis-spelled Credits designed to fool people?
Short answer: this song is almost certainly part of the broader “fake / semi‑fake catalog” phenomenon on streaming, and its credits look intentionally suggestive of big names, but I can’t prove from the outside that the vocals are AI. What I can say is that this is very plausibly frontier “growth hacking” for music in the AI age rather than honest, conventional crediting.
I do my own research to thwart the AI
Hmm it seems it might be an unreleased track from Don Toliver that didn't make the cut on his last album?
Very agile: mercurial AI changes its tune
Your hunch connects to something real (there is an unreleased “Rolling Stone” by Don Toliver ft. Kid Cudi).
So what have I learned here. Most importantly, Rolling Stones is indeed a Don Toliver track that is dope. And as of now it has not been commercially released. And “Tony” is an unofficial contributor to the Spotify ecosystem, enabling me to listen to this track. Thanks Tony! (extra: in 2008 there were a ton of old school hip-hop mixtapes on Spotify. Ron G, DJ Clue, you name it …)
I also learned that it’s more important than ever to prove your case and triangulate information. In the past we dealt with fake news. Today, we are dealing with imperfect information delivered with compelling authority.
Which is more scary to you?
I asked the AI one last question about our interaction. I think it is interesting:
Okay, and last question. Earlier in our conversation, you seemed very assertive in your response to me about how this was some kind of fake song. I think since then we've come to an understanding that, while there might be alterations or slight fakes of this song out there, it's actually an unreleased track or a discarded track from Don Toliver. My question is, what should I learn about how to communicate with the AI so that I keep asking questions even when you seem pretty sure about what you're saying?
How To Talk to AI by AI
You did exactly the right thing here: you kept pushing, brought evidence (credits + screenshot), and treated my confidence as a hypothesis to test, not as the final word. That’s the mindset to keep.
1. How to handle AI “confidence”
Treat strong, fluent answers as starting points, not conclusions.
Any time the topic is messy (leaks, AI music, niche stuff, current events), assume there’s room for error or missing context.
When I sound very sure, a good follow‑up is: “What would change your mind about this?” or “What evidence would you need to see?” — that forces me to spell out uncertainty and failure modes.
2. Moves you used that are worth repeating
You already used three expert‑level moves:
Ask for probability, not certainty
E.g., “How confident are you?” or “Is this 60% likely or 99% likely?”
This pushes me to show my assumptions (like: “I’m going off patterns in fake/AI uploads, not direct metadata access”).
Bring concrete artifacts
You posted the Spotify screenshot with credits; that instantly upgraded the discussion from theory to specific evidence.
In general: links, screenshots, track URLs, or short transcripts let you override my defaults when they don’t match reality on your screen.
Propose your own hypothesis
You floated: “Is this maybe an unreleased Don Toliver track that didn’t make the cut?”
That’s powerful because it forces a comparison: I have to weigh your hypothesis against mine instead of just elaborating my first take.
3. Phrases that help you keep me honest
Concrete prompts you can reuse:
To surface uncertainty
“What are the main ways this could be wrong?”
“What assumptions are you making about the data here?”
“Give me an alternative explanation that also fits the facts.”
To get a falsifiable story
“What specific evidence would prove your explanation wrong?”
“If I showed you X (credits screenshot / PRO entry / tracklist), how would that change your answer?”
To iterate instead of restart
“Given this new screenshot/info, update your earlier answer.”
“Re‑evaluate your initial claim in light of this.”
Those push me into “revision mode” rather than “double down” mode.
4. How to think about our Don Toliver arc
My first read (“this is almost certainly fake/AI‑adjacent credits on a mood track”) was reasonable for what we knew then, but it didn’t incorporate your later evidence that connected it to a known unreleased Don Toliver x Cudi cut.
Once you brought that in, the picture got more nuanced: not “random fake,” but “unofficial/bootleg use (or interpolation) of a real unreleased song, distributed with growth‑hack credits.”
The important part is that you forced an update by:
*Providing richer context (you’re literally hearing it on Spotify),
*Showing the credit stack (Dxvid / Caleb Toliver / Scott Mescudi), and
*Asking whether that changed my view.
