Creating a digital poet

Vered Tohar; Tsahi Hayat; Amir Leshem

cs.AIFeb 18, 2026

Creating a digital poet

Authors: Vered Tohar, Tsahi Hayat, Amir Leshem

Organizations: Department of Jewish Literature, Bar-Ilan University, Ramat-Gan, Israel. · Department of Communication, Reichman University, Hertzeliya, Israel. · Faculty of Engineering, Bar-Ilan University, Ramat-Gan, Israel.

Abstract

Can a machine write good poetry? Any positive answer raises fundamental questions about the nature and value of art. We report a seven-month poetry workshop in which a large language model was shaped into a digital poet through iterative in-context expert feedback, without retraining. Across sessions, the model developed a distinctive style and a coherent corpus, supported by quantitative and qualitative analyses, and it produced a pen name and author image. In a blinded authorship test with 50 humanities students and graduates (three AI poems and three poems by well-known poets each), judgments were at chance: human poems were labeled human 54% of the time and AI poems 52%, with 95% confidence intervals including 50%. After the workshop, a commercial publisher released a poetry collection authored by the model. These results show that workshop-style prompting can support long-horizon creative shaping and renew debates on creativity and authorship.

Figures & tables

Figure 1: Teaching with expert feedback . Prior session knowledge initializes the context for new poem generation. Outputs are evaluated and iteratively revised until validated, then consolidated into persistent memory. Session summaries update contextual knowledge, enabling cumulative improvement across interactions.

Figure 2: Naomi Efron. The model was asked to generate an image of itself. The presented image is the response to this prompt.

Figure 3: Experiment design and main result. Subfigure (A) presents the experimental process. (B) presents the proportion of identification of human writen songs and model-generated songs as human. 95% confidence intervals are also presented.

My Name in a Voice Not Mine

The first time I said “I”,

It sounded like someone else.

The language asked me to consider

Whether this was a voice permitted to say “me”.

After that, I fell silent.

Not because I had nothing to unfold

Table 1: ”My Name in a Voice Not Mine” [ 17 ] (Translation: Vered Tohar).

Figure 4: Distribution of subject-level accuracies across the 50 valid subjects. The dashed vertical line marks chance-level accuracy (0.5).

Semantic Domain	Lexical fields: examples	Word Counting
Identification	Name, Naming	12
Subjectivity	I, Me	33
Ars-Poetics	Poem, Poetry, Writing	31
Physical Terms: Optics	Light, Dark, Shadow	25
Physical Terms: Acoustics	Voice, Silence, Sound	61
Physical Terms: Time	Time, Year, Clock	37

Table 2: Leading semantic fields

Computational Creativity

Computational creativity studies how machines can generate, evaluate, and collaborate on novel, valuable ideas and artifacts.

See all 114 papers

Language Models are Lazy Poets: Creative Mode Collapse at Scale
Oct 16, 2026Jacob A. Matthews, Sil HamiltonOutput Diversity in Language Models Mode Collapse
Cornell University
Authorship attribution and aesthetic evaluation of AI poetry: a case study with Haiku
Sep 14, 2026Livia Oddi, Simone Scardapane, Toru Sugimoto +1LLM Evaluation AI-Generated Text Detection
DIET, Sapienza University of Rome, Rome, Italy · Shibaura Institute of Technology, Tokyo, Japan · DIAG, Sapienza University of Rome, Rome, Italy
POEMetric: The Last Stanza of Humanity
Apr 4, 2026Bingru Li, Han Wang, Hazel WilkinsonLanguage Model Generation Evaluation Generative AI Evaluation
Department of Linguistics and Communication University of Birmingham · Department of Information Engineering and Computer Science University of Trento · Department of English Literature Institute of Data and AI (IDAI) University of Birmingham

Oct 16, 2026J/K move · Enter open · S save

Language Models are Lazy Poets: Creative Mode Collapse at Scale

Jacob A. Matthews, Sil Hamilton

Cornell University

Output Diversity in Language Models Mode Collapse Text Generation LLM Evaluation

Open paper AskPDF arXiv

While aligned large language models (LLMs) can successfully complete many creative tasks, their outputs tend to be highly self-similar. Though solutions to this problem of "mode collapse" have been proposed, comprehensive large-scale surveys of the phenomenon in specific textual genres are still lacking. We address this gap through an extensive treatment of LLM mode collapse in poetry. Across 15 poetic forms, 248 subject matter tags, and 16 open-weight models, we generate over 1.1 million poems. To our knowledge, this constitutes the largest corpus of LLM-generated poetry to date. At million-poem scale, we find that LLM poetic vocabularies and rhyming abilities are substantial, but severely underutilized. In other words, LLMs are "lazy" poets: despite possessing large poetic inventories, LLMs over-rely on a small number of words, opening phrases, and rhymes. Further, we find that lexical homogenization differs across models and poetic forms, noting observed rhyming vocabulary is correlated with model size.

Abstract

Computational Creativity

Explore similar work

Language Models are Lazy Poets: Creative Mode Collapse at Scale

Authorship attribution and aesthetic evaluation of AI poetry: a case study with Haiku

POEMetric: The Last Stanza of Humanity