When a company releases a new Large Language Model AI product, it won't be long until someone puts in to the test, and benchmark it with other similar tools out there.
The benchmark include. and not limited to:
- Language Understanding and Generation – Text completion, summarization, translation, and paraphrasing.
- Knowledge and Reasoning – Question answering, commonsense and mathematical reasoning, code generation.
- Reading Comprehension – Extracting insights, multi-hop reasoning, and ensuring logical consistency.
- Bias and Fairness – Detecting and mitigating biases, ensuring equitable model outputs.
- Creativity and Adaptability – Storytelling, poetry, and role-based interactions.
- Robustness and Security – Testing against adversarial inputs and error recovery.
- Multimodal Capabilities – Handling images, video, and audio alongside text.
- Efficiency and Scalability – Measuring inference speed, memory usage, and performance at scale.
- Real-World Applications – Evaluating practical uses like dialogue systems, medical, and legal tasks.
But somehow, somebody devised a way to make this benchmarking a lot more fun, and hilarious.
And that is by creating a 'unofficial' benchmark, where various LLM models are told to create Will Smith eating spaghetti.
Apparently, results are indeed meme-worthy.
Due to the 'weird' AI benchmark and the trends that followed, and the meme that happened after that, Will Smith himself cared to join the crowd.
On February 2024, Will Smith shared a playful video on his official Instagram account, parodying the viral AI-generated clip of him eating spaghetti.
The video, captioned "This is getting out of hand!", uses a split-screen format.
At the top, labeled "AI Video 1 year ago," is the original AI spaghetti clip created by Reddit user “chaindrop” in March 2023. Below it, under “AI Video Now,” Smith appears in 11 segments, dramatically eating spaghetti—slurping, shaking his head, pouring it into his mouth with his hands, and even nibbling a friend’s hair.
Lil Jon’s 2006 hit Snap Yo Fingers plays throughout.
But best of this weird benchmark is that, it does show the capacity of LLM models throughout their development.
Since OpenAI introduced Sora, many people noted a dramatic jump in AI-video quality, when compared to the infamous, original Will Smith spaghetti video.
Since then, Will Smith and spaghetti becomes one of the most famous bizarre unofficial benchmark to take the AI community by storm.