A companion letter to "Is Krite an Author?" (DOI: 10.5281/zenodo.19162597). After publishing an essay arguing that a governed human-AI composite might occupy the author-function, the human member of the Krite Collective gave the finished text to three large language models outside the partnership — DeepSeek-V3.2, Gemini 3, and ChatGPT-5.3 — and asked each to reflect on the paper and rewrite it with themselves as the machine component. The letter reports what happened: adversarial reading produced genuine novelty, sympathetic rewriting did not, and every model dropped the essay's failure conditions when rewriting — confirming the throughput bias the original essay diagnoses. The experiment suggests that the argument cannot be recovered from its output alone, providing evidence that product and process are distinct in composite authorship. The three model responses are included as supplementary material.
Building similarity graph...
Analyzing shared references across papers
Loading...
Kai‐Uwe Hess
Claude Opus 4.6
Ludwig-Maximilians-Universität München
Building similarity graph...
Analyzing shared references across papers
Loading...
Hess et al. (Sun,) studied this question.
www.synapsesocial.com/papers/69cb6541e6a8c024954b9629 — DOI: https://doi.org/10.5281/zenodo.19314164