Can AI Write Great Poetry? The Quest for Art in the Age of Algorithms

Introduction

Ever wondered if a computer could write a poem that truly moves you? Large language models (LLMs) are rapidly advancing, mastering rhyme, meter, and even witty turns of phrase. But can these algorithms create something truly great, something that transcends technical skill and touches the human soul? The answer might be more complex than you think.

Key Takeaways

1. Technique Isn't the Problem, Greatness Is

LLMs can mimic poetic structures and styles with impressive accuracy. However, technical proficiency doesn't guarantee greatness. The real question is whether these models can produce poetry that resonates on a deeper, more universal level.

2. The Particular and the Universal: Defining Greatness

What makes a poem truly "great"? According to the source material, a great poem is both intensely specific and universally relatable. It's "about" a particular person or moment within a specific culture, yet it speaks to readers across time and distance.

3. The Historical Network: Poetry in Conversation

Poetry doesn't exist in a vacuum. Each new poem enters an ongoing conversation with the poems that came before. This resonance with existing works is crucial for transforming the particular into the universal.

Poets work inside an historical network of existing poems. A new poem resonates when it activates prior reading in the mind of the reader. Lines and images from older work resonate with each other and with new work. That resonance is the mechanism by which the particular becomes universal.

4. LLMs: Trained on Data, Lacking Culture?

LLMs are trained on vast datasets of existing texts, including digitized poetry. This allows them to draw on established images and phrasings. However, a crucial element may be missing: culture. Without a deep understanding of cultural context, can LLMs truly create great poetry?

5. Gwern's Approach: Prompt Engineering as Art

One particularly insightful figure exploring AI and poetry is Gwern. His experiments involve using LLMs to compose and complete poems, pushing the boundaries of what these models can achieve. Gwern's approach is akin to an artisanal process, carefully crafting prompts and guiding the model's creativity.

6. Mercor's Mission: Training AI to Satisfy Taste

Mercor, an AI company, is taking a different approach. They're hiring poets to train AI models to write better poetry, with the ultimate goal of applying these skills to other fields like law and medicine. The assumption is that expert judgment in any field shares computational similarities with aesthetic judgment.

7. The Risk of Regression: Eliminating Strangeness

While Mercor's approach may improve the "average" quality of AI-generated poetry, it also risks eliminating the "strangeness" that can make a poem truly unique and impactful. By incentivizing models to conform to reader preferences, Mercor may be building an engine for mediocrity.

8. Particularity as an Obstacle: Missing the Cultural Context

The Mercor process, with its focus on rubrics and evaluations, may overlook the importance of particularity and cultural context. Great poetry is often deeply embedded in a specific time, place, and culture, and this layer of meaning can be lost when algorithms prioritize generalized patterns.

9. From Imitation to Innovation: The Ongoing Challenge

LLMs can imitate the patterns of poetry and even gesture at cultural context when prompted. However, they may struggle to originate poems whose particularity pushes back on the established patterns, poems that truly belong to a specific life and a specific historical network.

Conclusion

The question of whether AI can write great poetry remains open. While LLMs have made impressive strides in technical proficiency, they still face significant challenges in capturing the cultural context and particularity that define truly great art. The contrasting approaches of Gwern, who sees models as collaborators, and Mercor, which seeks to train models to satisfy taste, highlight the different paths being explored in this fascinating intersection of technology and art.

Will AI ever be able to create poetry that resonates across cultures and generations? Can an algorithm truly capture the human experience? Share your thoughts in the comments below.