analiza aiqbench
This study provides critical insights into the internal mechanics of LLMs by demonstrating how specific neurons drive legal reasoning while others remain task-specific or universal across domains. The findings challenge assumptions about neuron layer distribution, showing that influential units depend heavily on input format rather than being universally concentrated in middle MLP layers. For researchers investigating model interpretability and domain adaptation, this work offers a rigorous framework for isolating genuine specialized knowledge from general capabilities within large language models.
Neuron Level Analysis of Large Language Model in Legal Domain Reasoning
arXiv:2606.15884v1 Announce Type: new Abstract: We presented a neuron-level analysis of legal-domain reasoning in LLMs, comparing it with other applied domain tasks across seven open-weight models. Using neuron attribution scores to rank and suppress influential neurons, we confirmed that suppressing the identified neurons collapses accuracy on the target task, whereas suppressing the same number of random neurons does not. We further found a small subset of neurons influential across all seven tasks; once these are removed, suppressing the remaining neurons degrades only the task they were identified from, revealing genuinely task-specific neurons in every model studied. Within the legal domain, the three benchmarks exhibit relatively high neuron overlap and tend to be affected jointly, suggesting of legal components neurons that span jurisdictions. The distribution of identified neurons in our experiments suggests that the hypothesis that influential neurons are concentrated in middle MLP layers may depend on the input format and content, rather than being a universal phenomenon.
Abstract:We presented a neuron-level analysis of legal-domain reasoning in LLMs, comparing it with other applied domain tasks across seven open-weight models. Using neuron attribution scores to rank and suppress influential neurons, we confirmed that suppressing the identified neurons collapses accuracy on the target task, whereas suppressing the same number of random neurons does not. We further found a small subset of neurons influential across all seven tasks; once these are removed, suppressing the remaining neurons degrades only the task they were identified from, revealing genuinely task-specific neurons in every model studied. Within the legal domain, the three benchmarks exhibit relatively high neuron overlap and tend to be affected jointly, suggesting of legal components neurons that span jurisdictions. The distribution of identified neurons in our experiments suggests that the hypothesis that influential neurons are concentrated in middle MLP layers may depend on the input format and content, rather than being a universal phenomenon.
Computation and Language (cs.CL)
arXiv:2606.15884 [cs.CL]
(or arXiv:2606.15884v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.15884 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.
Text integral sub licența deschisă a publicației · 247 cuvinte
Subiecte conexe
Harnessing the Reasoning Economy: A Survey of Efficient Reasoning for Large Language Models
arXiv · 7 sept. 2026
Controllable-CoT leads to covert reasoning capabilities
LessWrong · 22 sept. 2026
X-RAY: Mapping LLM Reasoning Capability via Formalized and Calibrated Probes
arXiv · 3 iun. 2026
Unsupervised Skill Discovery for Agentic Data Analysis
arXiv · 6 iun. 2026