Memory-Grounded Social Dynamics in Repeated LLM Agent Simulations: Dialogue-Only Transcript Supplement and Behavioral Evaluation Artifacts

Fuente: Zenodo
Enregistré dans:
Détails bibliographiques
Auteur principal: Ubaydullaev, Okhunjon
Format: Recurso digital
Langue:anglais
Publié: Zenodo 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866902087284031488
author Ubaydullaev, Okhunjon
author_facet Ubaydullaev, Okhunjon
contents <p>This record contains supplementary research artifacts for the preliminary behavioral evaluation report “Memory-Grounded Social Dynamics in Repeated LLM Agent Simulations”.</p> <p>The uploaded materials support qualitative inspection of repeated multi-agent LLM simulation outputs without disclosing the proprietary system architecture. The main supplement is a dialogue-only by-tick transcript package containing 70 full-batch main runs across seven model families and 6 diagnostic Gemini 3.1 runs. Transcripts are rendered as tick-level action/dialogue records and intentionally exclude prompts, hidden system instructions, memory-routing internals, parser rules, scoring thresholds, and proprietary implementation details.</p> <p>The purpose of this record is behavioral review rather than code-level reproducibility. The files allow readers to inspect generated scene behavior, action status, actor/target structure, public consequence text, and run-level coverage while preserving the closed-source nature of the simulation system.</p> <p>The associated report does not claim that agents are conscious, possess real inner lives, or exhibit proven identity transformation. The narrower claim is that persistent memory combined with social context can produce measurable, trajectory-sensitive social mechanisms across repeated LLM simulations, with outcomes and behavioral textures varying by model family.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19937581
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Memory-Grounded Social Dynamics in Repeated LLM Agent Simulations: Dialogue-Only Transcript Supplement and Behavioral Evaluation Artifacts
Ubaydullaev, Okhunjon
LLM agents
multi-agent simulation
social simulation
agent memory
generative agents
behavioral evaluation
dialogue transcripts
computational social simulation
narrative simulation
<p>This record contains supplementary research artifacts for the preliminary behavioral evaluation report “Memory-Grounded Social Dynamics in Repeated LLM Agent Simulations”.</p> <p>The uploaded materials support qualitative inspection of repeated multi-agent LLM simulation outputs without disclosing the proprietary system architecture. The main supplement is a dialogue-only by-tick transcript package containing 70 full-batch main runs across seven model families and 6 diagnostic Gemini 3.1 runs. Transcripts are rendered as tick-level action/dialogue records and intentionally exclude prompts, hidden system instructions, memory-routing internals, parser rules, scoring thresholds, and proprietary implementation details.</p> <p>The purpose of this record is behavioral review rather than code-level reproducibility. The files allow readers to inspect generated scene behavior, action status, actor/target structure, public consequence text, and run-level coverage while preserving the closed-source nature of the simulation system.</p> <p>The associated report does not claim that agents are conscious, possess real inner lives, or exhibit proven identity transformation. The narrower claim is that persistent memory combined with social context can produce measurable, trajectory-sensitive social mechanisms across repeated LLM simulations, with outcomes and behavioral textures varying by model family.</p>
title Memory-Grounded Social Dynamics in Repeated LLM Agent Simulations: Dialogue-Only Transcript Supplement and Behavioral Evaluation Artifacts
topic LLM agents
multi-agent simulation
social simulation
agent memory
generative agents
behavioral evaluation
dialogue transcripts
computational social simulation
narrative simulation
url https://doi.org/10.5281/zenodo.19937581