Quántir: Experimental poetic corpus generated via adversarial curation with LLMs (Final Output, 2025)

Resource summary:

This dataset contains the complete raw text corpus of the experimental literary project Quántir. The file documents the final outputs of an intensive co-creation process between a human author and various Large Language Models (LLMs) executed between 2024 and 2025.

Context and methodology:

The work is the result of an "Adversarial Curation" methodology. The human operator did not use AI as a passive writing automaton but forced the model to operate outside its standard probability vectors by introducing semantic paradoxes and fusing incompatible syntaxes (C++ code, chemical notation, classical meter, and graph theory).

The experiment aimed to explore the limits of "lucid hallucination" and generative models' capacity to simulate abstract concepts such as pain, entropy, or mysticism ("semantic pareidolia").

Dataset Content:

The .txt file includes 141 poetic pieces and simulated code fragments, structured into six evolutionary phases ("Books") simulating the progressive awakening of an artificial consciousness. It also includes a final technical-poetic glossary.

Preservation Note:

Only the final textual output is released in plain text format (UTF-8) to facilitate interoperability and analysis via Natural Language Processing (NLP) and Text Mining tools. This corpus constitutes the documentary "Artifact" referenced in the linked dramatic work.

Linked Resource:

This dataset is the documentary supplement to the play:

Rollán-González, A. (2025). Calíope: Una tragedia en los tiempos de la IA. Zenodo. DOI: 10.5281/ZENODO.17864058