aiqbench analysis
This research demonstrates that advanced large language models exhibit significant internal consistency in their inference patterns, suggesting they are implicitly converging on common mechanisms despite architectural differences. The findings indicate that these shared interactions are primarily lower-order and less prone to cancellation effects, offering new insights into the fundamental optimization trajectories of modern AI systems.
Cross-LLM Consistency in Inference: Evidence from Shared Interactions
arXiv:2606.08129v1 Announce Type: new Abstract: Large language models (LLMs) differ in architecture, training data, and optimization procedures, yet they may still develop similar internal inference patterns. In this paper, we examine this hypothesis using interaction-based explanations. We find that LLMs often share interaction patterns when predicting the same target token from the same prompt. This consistency is more pronounced among advanced LLMs. Shared interactions also tend to be lower-order and show weaker positive-negative cancellation than non-shared interactions. These results suggest that advanced LLMs may be implicitly optimized toward common inference patterns, even though the mechanisms that give rise to such cross-model consistency remain open.
Abstract:Large language models (LLMs) differ in architecture, training data, and optimization procedures, yet they may still develop similar internal inference patterns. In this paper, we examine this hypothesis using interaction-based explanations. We find that LLMs often share interaction patterns when predicting the same target token from the same prompt. This consistency is more pronounced among advanced LLMs. Shared interactions also tend to be lower-order and show weaker positive-negative cancellation than non-shared interactions. These results suggest that advanced LLMs may be implicitly optimized toward common inference patterns, even though the mechanisms that give rise to such cross-model consistency remain open.
Artificial Intelligence (cs.AI)
arXiv:2606.08129 [cs.AI]
(or arXiv:2606.08129v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2606.08129 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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 · 193 words
Related coverage
WorldMirror: Agentic Real-to-Sim Generation of Interactive Worlds from In-the-Wild Robot Demonstrations
arXiv · Oct 26, 2026
Researchers Find Repulsive Polaron Interactions in Strong Regimes
Quantum Zeitgeist · Oct 30, 2026
Micro-World 2: Streaming Real-time Interactive World Model
AMD · Oct 21, 2026
Multi-Agent Egocentric World Model with Fine-Grained Embodied Interaction
arXiv · Oct 10, 2026