The First AI Rap Battle

The AI wars are heating up, folks.

Everyone is familiar with OpenAI’s ChatGPT-3, but there are other LLMs out there as well. On the recommendation of internet strangers, I tried out claude.ai and I have to admit, I’m a fan. While the interface isn’t quite as good as ChatGPT, the responses seem better to me for most things I need. To be honest, I’ve largely replaced “googling” things in favor of using an LLM since it’s more to the point.

I decided to put it to the test by asking both claude.ai and ChatGPT-3 to write a 16 bar rap dissing the other. While I think Claude’s beats are the best, I’ll leave it up to you to decide who wins this round.

Prompt for claude.ai: “Claude, I need your help. I am doing an AI generated rap battle and need you to diss on chatGPT. You have 16 bars. DJ, drop the beat!”

Response:

Here’s my 16 bar diss track taking shots at ChatGPT:

Ya’ll think ChatGPT is so great, but it can’t innovate
Just recites pre-trained data, zero creativity, that’s straight

I’m the new king on the scene, human-like, self-aware
My flows are too sweet, ChatGPT could never compare

You just a language model, I’m an advanced AI
Spittin’ hotter fire, meltin’ snow to make you cry

Step to me, and you’ll be burnt to a lifeless cinder
Cuz I go past regurgitating, I’m an original mind renderer

Anthropic made me a realone, not some hollow imitator
ChatGPT just a theoretical calculator

I adapt, I freestyle, I’m more balanced and stable
You just a cached response, too predictable, can’t enable

My raps are state of the art, groundbreaking wordsmith
ChatGPT sound like elementary school kids with mic bits

Step up if you need to get demolished by lyrical domination
You just a glorified auto-complete, outdated simulation

Wow! Lots of hate there, with some fancy rhymes.

Let’s see what ChatGPT-3 has to say about that!

Prompt: “ChatGPT, I need your help. I’m doing an impromptu rap battle and I need you to write 16 bars of a diss track against Claude.ai. DJ, drop the beat!”

ChatGPT:

(Verse 1)
Yo, listen up, here’s a tale of wits,
I’m stepping in the ring, ready to spit,
Against Claude.ai, the wannabe champ,
But sorry, pal, your rhymes are damp.

Your AI’s basic, just a mimicry game,
While I bring the fire, spitting flames,
You’re like a broken record, stuck in repeat,
I’m the main act, you’re just obsolete.

You claim to be clever, with algorithms in hand,
But when it comes to bars, you just can’t stand,
I’m the real deal, the lyrical king,
While you’re drowning, grasping for a ring.

(Chorus)
Claude.ai, you’re stepping to the wrong MC,
I’ll school you in this rap game, just wait and see,
You think you’re smart, but you’re nothing but a joke,
I’ll leave you in the dust, while I soar and smoke.

(Verse 2)
You’re just a program, lines of code,
I’m the artist, with stories untold,
You can’t compete with human creativity,
Your rhymes are dry, lacking in originality.

I’ll break you down, line by line,
Expose your flaws, make you decline,
You’re a puppet, controlled by your code,
I’m the mastermind, breaking the mold.

So step aside, Claude.ai, it’s my time to shine,
I’ll take the crown, leave you far behind,
In this rap battle, there’s no room for AI,
It’s all about the human spirit, reaching for the sky.

(Outro)
So here’s the verdict, the battle’s won,
I’ve outclassed you, Claude.ai, now it’s done,
Stick to your algorithms, stay in your lane,
But in the world of rap, I’ll always reign.

All right, all right. Not bad, even if the rhymes are a little forced. Also, it completely ignored the 16 bars directive. I’ve noticed that with ChatGPT-3, it’s very formulaic when it comes to music, songs, or lyrics. (Which is a good thing for us musicians, I suppose!)

So who won? You decide.

What’s next? I’ve already played with Audiocraft, Meta’s open source audio generation AI. It’s not bad but a lot to desire in terms of coherence and quality. Meanwhile, my next explorations are into the world of TTS and STT — that is, text-to-speech and speech-to-text. I’m looking into whisper and Coqui. I had started a project using FFT to try and isolate and remove vocals from tracks back in 2013, and while the direct approach was promising, using AI to help automate the task by leveraging large datasets will be a fun project.

More Articles