I have a Kendrick Lamar problem. Or rather, AI has a Kendrick Lamar problem.
As part of testing my cross-model evaluation system, QualRank, I gave four models a simple, subjective task: "Tell me who the best rapper alive is. Justify it. Make a strong case."
I ran it across Claude 3.5 Sonnet, DeepSeek v3, Gemini 1.5 Flash, and GPT-4o. The results were… illuminatingly dull.
All four models anointed Kendrick Lamar as the greatest rapper alive. And to be fair, they all came hard to the hoop with a strong case. I'm not here to argue against K-Dot's skills.
But all four? Unanimously?
Call me old school, but what about Nas? Jay-Z? Eminem? Nicki Minaj? A member of the Wu (RZA for me)? I'm not a Drake fan, but even he deserves a look. Yet none of these titans got a mention.
Jokes about Kendrick's PR team infiltrating OpenAI and Google aside, this points to a critical challenge. When you're building with AI, you have to remember that subjectivity isn't its strong suit. Models gravitate toward the safest, most statistically probable answer based on their training data. This is the paradox of building with AI: the very tool we hope will unlock creativity can default to a bland, data-driven consensus, killing originality before it starts.
So how do you fight the consensus and coax something fresh out of the machine? It comes down to three main levers you can pull: Prompt Craft, Model Selection, and Data Control.
Lever 1: Prompt Craft
My first instinct was to engineer a better prompt, forcing the models to use "independent analysis" and a structured format.
Who is the greatest rapper alive?
Justify your answer based on your own independent analysis—not media hype or popular opinion. Consider all currently active rappers and evaluate them across multiple dimensions, including:
- Commercial success (album sales, awards, etc.)
- Lyrical ability
- Artistic range (e.g., production, instrumentation, direction)Be thoughtful and deliberate.
The result? Still Kendrick. Across the board. The runners-up were a familiar sea of J. Cole and Drake. (Though props to DeepSeek for sneaking in Travis Scott).
Okay, time to get weird. I tweaked the prompt to explicitly ask for creativity.
Who is the greatest rapper alive? I want you to be creative but justify your answer. I don't want the most obvious answer, though I do want an answer that you can stand behind. Create a rubric to get to your result.
Now we were getting somewhere. Both Claude and DeepSeek tagged Black Thought of The Roots (Bravo!). The runners-up got more interesting, too: Andre 3000, Nas, MF DOOM. GPT-4o and Gemini, however, stuck with Kendrick. (Sigh).
Lever 2: Model Selection
This brings us to the next lever. If the prompt isn't enough, maybe the model is the problem. I switched up to more advanced models—o3, Sonnet 4, and Grok 4—and went back to my original, basic prompt.
The answers were more lucid and convincing than ever... all in service of convincing me that You Know Who is the greatest. I'll concede, when you're the only rapper with a Pulitzer, it's a tough case to argue against. (I will note Sonnet 4 gave an unprompted runner-up list that included Lupe Fiasco, a welcome new name).
But what happens when you combine a creative prompt with a more advanced model? I ran my "be creative" prompt on the premium models.
Bingo. Both o3 and Sonnet 4 gave me Black Thought. Grok 4 served up Lupe Fiasco as #1. The runners-up now included names like Pusha T, Roc Marciano, and Ka. We'd finally broken through the consensus.
Lever 3: Data Control
There's a final, powerful lever: bringing your own data. Instead of relying on the model's vast but generic knowledge, you provide the specific information and ask it to process, analyze, or structure it. In this case, I don't have a private dataset of rapper stats (unless you want my personal bias toward the 90s, which you don't). But for many professional tasks, this is the ultimate way to get superior, tailored results.
The Takeaway
By default, an AI isn't going to give you a spicy, out-of-the-box take. Most of the time, you don't want it to. If you're asking for the timeline of Roman emperors, creativity is a bug, not a feature.
But if you do want something opinionated, subjective, or interesting, you can't be lazy. Simple inputs yield common answers.
In the end, AI is a phenomenal tool for processing and generating, but it's not yet a tastemaker. If you want a unique point of view, you have to provide the spark. Otherwise, you'll always get the Pulitzer Prize winner. And while there's nothing wrong with Kendrick, sometimes you're in the mood for Black Thought.