Work

Ammons LLM Preamble

Year: 2026
Type: article
Abstract: A model can use a work without quoting it. Ammons LLM Preamble defines an operational measure of the resulting behavioral change under controlled exposure and proposes a falsifiable evidence-pooling model. For a declared query population, the Ammons Influence Probability is the probability that a semantic response-distribution shift exceeds a chosen threshold. The paper derives conditional scarcity and dose results, invariance properties, robustness bounds, and finite-sample measurement bounds. Its training section is analytical and distinct from the encounter-time experiment. A separate pilot evaluates 12 paired instances of one newly generated rule family in 24 fresh contexts, with two answers per context. Relevant work exposure produces 24/24 correct answers; matched irrelevant content produces 0/24. The paired instance is the analysis unit. The result demonstrates context-time use of supplied rules. It does not validate the pooling law, identify hidden training sources, or establish persistent learning. The public materials include the paper, editable LaTeX, generated protocol, retained responses, and Python standard-library programs for exact mathematical checks and rescoring. Offline reproduction of the retained analysis does not generate new model evidence. I invite technical criticism of the measurement design and controlled tests of the model’s predictions. Public source and reproduction materials: 10.5281/zenodo.24177478
Author: Ammons, Dean
Language: English
Open Access status: green
OpenAlex ID: W7225916931
FWCI:
Cites:
Cited by:
Related to: 10