Researchers propose KDASO to automate and optimize legal liability adjudication skills for LLMs
KDASO addresses staleness, semantic execution gaps, and context budget limits in LLM skill libraries for liability adjudication tasks.
1 source · cross-referenced
- KDASO introduces a method to continuously evolve skill libraries for liability adjudication tasks by leveraging recent adjudication data and LLM trajectory analysis.
Researchers from multiple institutions propose Knowledge-Data Dual-Driven Automated Skill Optimization (KDASO), a framework designed to address three key limitations in using large language models (LLMs) for liability adjudication tasks such as legal charge determination.
The work highlights that existing approaches suffer from stale skill libraries as real-world scenarios evolve, semantic execution gaps where LLMs deviate from intended expert logic, and context budget constraints that lead to critical skill omissions as skill collections grow.
KDASO introduces a continuous optimization process that includes neighborhood-conditioned induction from recent adjudication data, execution-aligned refinement based on LLM trajectory analysis, and context-budgeted routing for scalable skill selection.
The authors evaluate KDASO across three domains: ride-hailing adjudication, legal liability determination, and logical reasoning judgment, reporting significant performance improvements over general-purpose and domain-specific baselines.
The framework achieves an 8.96% improvement when transferring skills to concept-similar scenarios and maintains consistent performance gains through continuous optimization, demonstrating high transferability and stability of the generated skills.
- Nov 11, 2026 · The Verge — AI
Google DeepMind unveils AlphaGenome 2 to predict genetic variant effects on cell types
Trust79 - Nov 11, 2026 · Google DeepMind — Blog
Google DeepMind unveils AlphaGenome 2 to predict DNA variant effects in specific cell types
Trust84 - Nov 11, 2026 · arXiv cs.AI
Researchers demonstrate auditable human-agent teams for autonomous synchrotron X-ray experiments
Trust79