Mutation Without Variation: Convergence Dynamics in LLM-Driven Program Evolution

A new study on convergence dynamics in LLM-driven program evolution reveals that large language models tend to cycle back to previously seen code structures rather than exploring novel solutions. This 'mutation without variation' phenomenon poses a significant challenge for using AI in autonomous software development.

Computer Science > Artificial Intelligence

Title:Mutation Without Variation: Convergence Dynamics in LLM-Driven Program Evolution

View PDF HTML (experimental)Abstract:When an LLM repeatedly mutates a program, does it explore new forms or circle back to the same ones? We study this question by analyzing LLM-driven mutation chains in the absence of selection pressure within a domain-specific language, varying prompt design, model family, and stochastic replication. We find that LLM-based mutation consistently converges toward restricted attractor regions in program space. Convergence is especially severe at the structural level: in 87% of chains, over 93% of mutations revisit a previously seen structural form, with most variation confined to terminal substitutions within recurring templates. Cycle analysis reveals short cycles and self-loops dominating the transition structure. The rate of convergence varies with prompt wording and model choice, but the phenomenon is robust across conditions. A classical GP subtree mutation operator does not exhibit comparable convergence, suggesting that the effect is intrinsic to the LLM mutation pipeline. These findings reveal a tension at the heart of LLM-driven program evolution: the same capabilities that enable semantics-aware program transformation also carry a systematic bias toward structural homogeneity that must be accounted for if such systems are to sustain open-ended exploration. Source code is available at this https URL.

Bibliographic and Citation Tools

Code, Data and Media Associated with this Article

Demos

Recommenders and Search Tools

arXivLabs: experimental projects with community collaborators

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.

Source: arXiv cs.AI Recent

More in this category

agentic-systems

Minimizing the Hidden Cost of Scales: Graph-Guided Ultra-Low-Bit Quantization for Large Language Models

Researchers have proposed SAGE-PTQ, a novel ultra-low-bit post-training quantization framework for LLMs that minimizes hidden scaling overhead. It significantly reduces GPU memory usage and accelerates decoding speed while maintaining high accuracy compared to existing methods.

agentic-systems

Agents' Last Exam

A new benchmark called "Agents' Last Exam" (ALE) has been introduced to evaluate AI agents on long-horizon, economically valuable, real-world tasks, revealing that current models achieve an average pass rate of just 2.6% on the hardest tier.

agentic-systems

A Motivational Architecture for Conversational AGI

This paper proposes a conversational reinterpretation of the OpenPsi motivational lineage, coupled to MetaMo's higher-level motivational scaffold, for agents built on a modular execution substrate.

agentic-systems

Assessing the Carbon Emissions and Energy Consumption of U.S. Hyperscale Data Centers

A new study analyzing 403 hyperscale data centers in the US reveals that the AI boom is driving electricity consumption and carbon emissions to alarming levels, with their carbon intensity averaging 48% higher than the national grid average.

NOW LET US Related – An interpretable and trustworthy AI framework for large-scale longitudinal structure-pain association studies using data from the Osteoarthritis Initiative (OAI)

agentic-systems

An interpretable and trustworthy AI framework for large-scale longitudinal structure-pain association studies using data from the Osteoarthritis Initiative (OAI)

Researchers have developed an interpretable and trustworthy AI framework to study the relationship between knee joint structural abnormalities and pain progression. By combining deep learning with advanced statistical modeling, this framework significantly improves prediction accuracy and clinical reliability.

agentic-systems

Residual Modeling for High-Fidelity Learned Compression of Scientific Data

Researchers have proposed a new residual modeling approach to optimize machine learning-based scientific data compression. This solution addresses the block-level accuracy degradation issue, outperforming traditional methods in high-fidelity scenarios.

EXPLORE TOPICS