Real Alpha
Watch mode

Lex Fridman Podcast · 5 min instead of 2 h 34 m · Mar 1, 2026

Why AI songs get spotted, and who benefits from AI

Rick Beato

Founder, Beato Club

Music educator, producer and songwriter whose YouTube channel on great musicians and music theory reaches millions of viewers.

  • Lex FridmanHost · Host, Lex Fridman Podcast

    Robotics and AI researcher at MIT; started the show in 2018.

About this segment

Beato explains his Suno workflow and why only strong songwriters can pick the good output from the bad. His kids spot AI instantly through reverb artifacts, which he traces to low-bitrate MP3 training data.

From “Rick Beato: Greatest Guitarists of All Time, History & Future of Music | Lex Fridman Podcast #492”, Mar 1, 2026. Watch the full episode

“the people that are already great songwriters, because you have to be able to recognize when it spits out something good.”
Rick Beatoat 1:47:13 · Lex Fridman Podcast1:47:13

Takeaways

  1. Lyrics from Claude, music from Suno; about three good songs per 130 ideas
  2. AI artifacts show up in vocal reverb and ambience
  3. Labels' multitrack deals will make the training data better
Watch 4:37 from 1:46:13

In this episode

Full episode

  1. 1:46:13Why AI songs get spotted, and who benefits from AI4:37

  2. 2:04:39Fighting YouTube copyright claims on fair use paid off2:14Up next
  3. 1:10:30Why the Beatles' best work came once they stopped touring2:44

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 1:46:20

    everybody chill calm down so if you write a prompt in suno and it spits out a song which i've did And I've made a bunch of videos in this. I made up a fake artist, Eli Mercer, in this video. Then I did a thing for CBS News. I made up this fake artist, Sadie Winters, and came up with this song, Walking Away. Well, the program came up with it. There is some creativity in a process. So in this particular thing, the process is you generate an image.

  2. 1:46:47

    I did it in ChatGPT, the image. Then I went to Claude, and I wrote the lyrics, because Claude's way better at lyrics than Suno is. Suno's bad at lyrics, at least right now. So I created the lyrics in Claude, and then I imported the lyrics into Suno. And I had great results with the songs that it came up with. I always have to qualify that. But I started thinking about this. People freak out about this. Oh, this is bad. This is bad. And then I was like, no, who are going to be the ones that are going to benefit from AI? Well, the people that are already great songwriters, because you have to be able to recognize when it spits out something good.

  3. 1:47:26

    Versus when it spits out something that's not that good. And every other song, I've probably created 130 song ideas, out of which there's three good ones. And there's a thing that's happening where people's ear very quickly is becoming attuned to AI slop. Yes. And that's actually quite fascinating. Like, for example, one of the things, there's this viral clip going around of an AI-based, like a soul jazz remix of songs like 50 Cent, Many Men. And I think it is super impressive. And it's a different pipeline, actually. It's a tricky pipeline how to pull that off. And I think a lot of the creativity in that, even that kind of remixing, is in the pipeline of how you actually do that.

  4. 1:48:12

    because there's actually a lot of manual stuff in that pipeline uh but i think ironically it's very cool at first but when you listen to it for a while you understand that this is ai slop yes for a soul remix it it actually lacks soul but it made me think of like when i listen to soul or blues i think i really want in that case to know i don't want to ai bb king i want the real bb king and i if i if i know if any ai is involved in the bb king process i'm tuning out yes and i don't think i'm being curmudgeonly old dude in that i think we humans want authenticity so when when AI, when I first started making these AI videos, it started back in 2023. I made my first one.

  5. 1:49:12

    And I would take my phone, come up in the kitchen, I'd play a song. And my youngest and Dylan, my youngest, Layla, and I have three kids, and my oldest, Dylan. As soon as I played, why are you listening to AI? And I was like, oh my God, instantly. I was like, how do you know? oh it has this ringing sound in the thing so it took me probably about four or five days to figure okay what are they hearing that i'm not hearing so i did it i separated all the parts and what they're hearing was the artifacts that are in the vocal reverb that sent that were uh that made incomplete it just couldn't do the ambiences correctly right because it's trained on a lot of these AI programs are trained on very low bit rate MP3s, right? So they feed all this stuff in there. So they're getting really inferior information in the training process. Whereas now

  6. 1:50:07

    when they make these deals with the major labels, they'll get the multi-tracks and they'll get high quality WAV files to train from, right? And whoever opts in, they get the solo vocal tracks, you know if Ed Sheeran wants to do it or Drake or whoever wants to give their voice to it let it do its thing and then get their royalties from it I'm not saying that any any of them are doing I'm just giving an example but every time that I would do it I could be down the hall and I would play something on my phone just to see if they'll like why you listen to AI oh they can instantly tell then eventually started getting better and then it'd be like is this AI I'd be in the car