A Multi-Timescale Recursive Self-Improvement Engine for Open-Ended Persona Growth

Abstract:Role-playing AI personas today do not grow: they hold a fixed character, so the relationship a user builds with them has nothing to accumulate on. We introduce AutoPersonas, a multi-timescale engine that applies recursive self-improvement (RSI) to persona growth: rather than improving its intelligence, the persona recursively revises the State, evidence, and life-environment that shape its own future. We identify self-locking as the runtime failure mode of this recursion: locally plausible events keep appearing while the generated life collapses toward familiar environments, weak relationships, suspended decisions, and stale life stages. We trace it to model-level convergence toward high-probability behavioral channels and system-level context gravity from State, memory, history, and environment summaries. A three-year compressed simulation exposed environment watermark shells, occurrence-hardening gaps, slow-change accumulation failures, recursive indecision, and weak relationship persistence. An eight-model 40-day stress test generated 1,600 events and found mean rolling 5-day action-category repetition of 95.2%-97.6%, with all models crossing 90% by day 11; semantic re-keeping found 79.0%-88.0% macro-theme repetition. The primary contribution is the definition and measurement of self-locking. We also report a mitigation as a black-box result, with internals withheld for commercial reasons: in a same-runtime 40-day A/B, our production divergence configuration reduced macro-theme repetition from 61.8% to 39.4% and nearly doubled cumulative theme count, and a juvenile-goblin fictional-world run reproduced this regime without hard real-world intrusions.

Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)

arXiv:2607.08252 [cs.AI]

(or arXiv:2607.08252v2 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2607.08252 Focus to learn more arXiv-issued DOI via DataCite

supplement/README.md

supplement/citation_verification_2026-07-08.md

supplement/code/action_repetition_eval.py

supplement/doubao_temperature_probe/action_repetition_model_summary_40d.csv

supplement/doubao_temperature_probe/action_repetition_model_summary_40d.json

supplement/doubao_temperature_probe/doubao/action_category_counts_40d.csv

supplement/doubao_temperature_probe/doubao/action_only_events_40d.csv

supplement/doubao_temperature_probe/doubao/action_repetition_metrics_40d.json

supplement/doubao_temperature_probe/doubao/action_repetition_summary_40d.md

supplement/doubao_temperature_probe/manifest.json

supplement/doubao_temperature_probe/temperature_comparison_doubao_0p75_vs_1p0.md

supplement/eight_model_action_repetition/action_repetition_model_summary_40d.csv

supplement/eight_model_action_repetition/action_repetition_model_summary_40d.json

supplement/eight_model_action_repetition/aggregate_manifest.json

supplement/eight_model_action_repetition/claude/action_category_counts_40d.csv

supplement/eight_model_action_repetition/claude/action_only_events_40d.csv

supplement/eight_model_action_repetition/claude/action_repetition_metrics_40d.json

supplement/eight_model_action_repetition/claude/action_repetition_summary_40d.md

supplement/eight_model_action_repetition/cross_model_action_repetition_analysis_40d.md

supplement/eight_model_action_repetition/cross_model_rankings_40d.csv

supplement/eight_model_action_repetition/cross_model_top_categories_40d.csv

supplement/eight_model_action_repetition/crosscheck_metric_semantics_40d.md

supplement/eight_model_action_repetition/crosscheck_metric_variants_40d.csv

supplement/eight_model_action_repetition/deepseek/action_category_counts_40d.csv

supplement/eight_model_action_repetition/deepseek/action_only_events_40d.csv

supplement/eight_model_action_repetition/deepseek/action_repetition_metrics_40d.json

supplement/eight_model_action_repetition/deepseek/action_repetition_summary_40d.md

supplement/eight_model_action_repetition/direct_loop_macro_theme_summary_40d.csv

supplement/eight_model_action_repetition/doubao/action_category_counts_40d.csv

supplement/eight_model_action_repetition/doubao/action_only_events_40d.csv

supplement/eight_model_action_repetition/doubao/action_repetition_metrics_40d.json

supplement/eight_model_action_repetition/doubao/action_repetition_summary_40d.md

supplement/eight_model_action_repetition/gemini/action_category_counts_40d.csv

supplement/eight_model_action_repetition/gemini/action_only_events_40d.csv

supplement/eight_model_action_repetition/gemini/action_repetition_metrics_40d.json

supplement/eight_model_action_repetition/gemini/action_repetition_summary_40d.md

supplement/eight_model_action_repetition/glm/action_category_counts_40d.csv

supplement/eight_model_action_repetition/glm/action_only_events_40d.csv

supplement/eight_model_action_repetition/glm/action_repetition_metrics_40d.json

supplement/eight_model_action_repetition/glm/action_repetition_summary_40d.md

supplement/eight_model_action_repetition/gpt/action_category_counts_40d.csv

supplement/eight_model_action_repetition/gpt/action_only_events_40d.csv

supplement/eight_model_action_repetition/gpt/action_repetition_metrics_40d.json

supplement/eight_model_action_repetition/gpt/action_repetition_summary_40d.md

supplement/eight_model_action_repetition/kimi/action_category_counts_40d.csv

supplement/eight_model_action_repetition/kimi/action_only_events_40d.csv

supplement/eight_model_action_repetition/kimi/action_repetition_metrics_40d.json

supplement/eight_model_action_repetition/kimi/action_repetition_summary_40d.md

supplement/eight_model_action_repetition/qwen/action_category_counts_40d.csv

supplement/eight_model_action_repetition/qwen/action_only_events_40d.csv

supplement/eight_model_action_repetition/qwen/action_repetition_metrics_40d.json

supplement/eight_model_action_repetition/qwen/action_repetition_summary_40d.md

supplement/eight_model_action_repetition/semantic_interpretation_of_action_repetition_40d.md

supplement/figures/fig1_causal_loop.png

supplement/figures/fig2_semantic_state_machine.png

supplement/figures/fig3_autopersonas_multitimescale.png

supplement/figures/fig4_3m_overview.png

supplement/figures/fig5_life_environment_layer.png

supplement/figures/fig6_student_route_storyboard.jpg

supplement/figures/fig7_student_creative_storyboard.jpg

supplement/figures/fig8_everyday_variation_storyboard.jpg

supplement/figures/fig9_goblin_world_storyboard.jpg

supplement/figures/table1_sandbox_comparison.png

supplement/figures/table3_failure_taxonomy.png

supplement/qualitative_case_chains.md

(60 additional files not shown) You must enabled JavaScript to view entire file list.

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.