aiqbench analysis
This arXiv paper studies what happens when LLM outputs are repeatedly fed back into the same model with no other context, modeling the process as a Markov chain. The finding that iterated rephrasing or round-trip translation either collapses to a small recurrent set of sentences or keeps generating novel text, with diversity rising or falling depending on temperature and the starting sentence, is a useful empirical result. For readers building multi-agent or pipeline systems, it signals that chained LLM calls can quietly lose or homogenize content, so diversity and drift should be measured rather than assumed.
Markovian Generation Chains in Large Language Models
arXiv:2603.11228v2 Announce Type: replace Abstract: The widespread use of large language models (LLMs) raises an important question: how do texts evolve when they are repeatedly processed by LLMs? In this paper, we define this iterative inference process as Markovian generation chains, where each step takes a specific prompt template and the previous output as input, without including any prior data. In iterative rephrasing and round-trip translation experiments, the output either converges to a small recurrent set or continues to produce novel sentences within a finite horizon. Through sentence-level Markov chain modeling and analysis of simulated data, we show that iterative process can either increase or reduce the diversity depending on factors such as the temperature parameter and the initial input sentence. These results offer valuable insights into the dynamics of iterative LLM inference and their implications for multi-agent LLM systems.
Abstract:The widespread use of large language models (LLMs) raises an important question: how do texts evolve when they are repeatedly processed by LLMs? In this paper, we define this iterative inference process as Markovian generation chains, where each step takes a specific prompt template and the previous output as input, without including any prior data. In iterative rephrasing and round-trip translation experiments, the output either converges to a small recurrent set or continues to produce novel sentences within a finite horizon. Through sentence-level Markov chain modeling and analysis of simulated data, we show that iterative process can either increase or reduce the diversity depending on factors such as the temperature parameter and the initial input sentence. These results offer valuable insights into the dynamics of iterative LLM inference and their implications for multi-agent LLM systems.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2603.11228 [cs.CL]
(or arXiv:2603.11228v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2603.11228 Focus to learn more arXiv-issued DOI via DataCite
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.
Full text under the publisher's open licence · 234 words
Related coverage
Non-normality and dissipation in Markovian quantum dynamics: Implications for quantum simulation
arXiv · Aug 5, 2026
Volterra Generative Models via Markovian Lifting
arXiv · Nov 6, 2026
Physics-informed traditional and generative Mori-Zwanzig-reduced models for multiscale systems
arXiv · Nov 5, 2026
The Trinity of Markovian Quantum Thermodynamics: Unifying the Axiomatic, Microscopic, and Operational Paradigms
arXiv · Jul 20, 2026