Beyond the beat: How AI interprets Lil' Wayne's lyricism

An AI parses evocative lyrics much differently from the way a human does

For reasons that espcape me, last night I was thinking about the HBO show Ballers, in which The Rock plays an ex-NFL player-turned-sports agent in Miami. Hilarity, replete with women in bikinis cavorting on yachts, ensues. It is, as you might imagine, Entourage with more testosterone and humidity. Replace Los Angeles with Miami, and, well, you get the picture.

Here’s the show’s opening:

The song is Lil’ Wayne’s Right Above It.

But this Substack is about AI, not pop culture. Bear with me: there is an AI angle to be had here, and it is an interesting one. When I first watched this show, I was struck by one of Lil’ Wayne’s lines: “Slumdog Millionaire Bollywood flow.” He’s co-opting the association between the movie Slumdog Millionaire and India (“Bollywood,” even though the movie was written by one British person and directed by another British person), and his own life’s story of having managed to escape poverty in New Orleans to party with the rich and famous in LA, Miami, Las Vegas, etc. The song’s name, Above It All, makes this explicit: he has risen above the systemic dysfunction that would have otherwise consigned him to penury.

What you just read, in the foregoing paragraph, is a human’s exegesis of this lyric. This is much different from an AI’s analysis. I was curious to see what ChatGPT would make of a prompt which simply read Slumdog Millionaire Bollywood flow. Here’s the exchange:

Me: slumdog millionaire bollywood flow

ChatGPT: The lyrics you’re referring to are from Drake’s “Bollywood Flow”. Here’s a snippet of the lyrics:

“This that Slumdog Millionaire Bollywood flow and ahh, my real friends never hearing from me fake friends write the wrong answers on the mirror for me.”

For the full lyrics, you can visit the source.

Me: How did you know that these were lyrics to a song?

ChatGPT: When you mentioned “Slumdog Millionaire Bollywood flow,” it resembled a line from a song. This is because phrases like this are often unique to song lyrics, especially when they include specific references such as “Slumdog Millionaire,” a well-known film, combined with “Bollywood flow,” which suggests a musical or lyrical element. Additionally, my training includes recognizing patterns in text that are characteristic of song lyrics, poetry, and other forms of creative writing. This recognition is what led me to search for the lyrics to confirm and provide you with the accurate information.

Note, of course, that ChatGPT makes two mistakes here, by claiming (1) that the song is Drake’s, and (2) that its title is “Bollywood Flow”. Neither assertion is correct; these are hallucinations. Drake did collaborate with Lil’ Wayne on this song, though the song most commonly seems to be associated with Lil’ Wayne, and not Drake.

But these hallucinations aren’t the interesting thing to think about. What is more interesting to think about is how a large language model (LLM) like ChatGPT parses text, and how that differs from how a human parses the same text. Parsing language, whether done by an LLM like ChatGPT or by a human, is fundamentally rooted in pattern recognition and contextual understanding. But the way LLMs do this is much different from the way that humans do it.

Here’s how this works with LLMs:

1. Pattern Recognition: LLMs rely on statistical patterns in text data. They identify and generate language based on probabilities derived from a vast collection of text. This process is devoid of understanding or consciousness.

2. Contextual Analysis: LLMs analyze context through adjacent words or phrases, but the comprehension is limited to the data they’ve been trained on. Further, this analysis lacks real-world awareness or experiences.

3. Limitations in Understanding: While LLMs can identify patterns that resemble song lyrics, they cannot appreicate or understand the cultural, emotional, or artistic significance of these lyrics.

Here’s how this works with humans:

1. Cognitive Understanding: Humans parse language with an innate cognitive ability. Humans understand not just the words used, but also their nuanced meanings, emotional underpinnings, and cultural contexts. Recall what I refer to as my (a human’s) exegesis of the line Slumdog Millionaire Bollywood flow, above.

2. Experiential Context: Humans’ understanding of language is enriched by personal experiences and cultural background, allowing for a deeper appreciation of references like “Slumdog Millionaire,” a well-know cultural artifact. This is not to suggest that my background is similar to Lil’ Wayne’s: it is not, at all. However, we both share knowledge of “Slumdog Millionaire” and the cultural meanings to which it refers.

3. Emotional and Artistic Interpretation: People appreciate the artistry in language, understanding and feeling emotions conveyed through phrases like “Bollywood flow.” LLMs can’t do this.

When I prompted ChatGPT with the lyric Slumdog Millionaire Bollywood flow, the LLM recognized it as a probable line from a song, based on its training in identifying patterns consistent with song lyrics. This identification was a result of a statistical likelihood rather than any understanding of the cultural or artistic significance of the phrase. In contrast, a human might recognize the phrase, understand its reference to the film “Slumdog Millionaire,” and appreciate the cultural fusion implied by “Bollywood flow.” This human interpretation would be enriched by personal experiences, cultural knowledge, and perhaps an emotional connection to the movie or music.

Understanding this discintion is critical. It turns out that understanding how ChatGPT parses language, as compared to how a human parses the same language, is helpful in understanding why ChatGPT responds in the manner it does. I noted above that ChatGPT hallucinated in its response to my query about how it knew that the prompt I gave it was a song lyric. ChatGPT has no sense, in the way that a human does, of who Drake or Lil’ Wayne is, or what a song is, or whether “Bollywood Flow” is the name of a song. LLMs in general, and ChatGPT in particular, is a much different kind of intelligence than the one that humans use.