Three New arXiv Papers Explore LLM Optimization, Healthcare Applications, and Psychological Analysis

Three recent papers published on arXiv address different challenges in AI systems:

MultiGA: Genetic Algorithms for LLM Refinement

According to arXiv paper 2512.04097v1, researchers propose using evolutionary algorithms inspired by natural selection to refine Large Language Model performance. The paper notes that “LLMs are widely used across research domains to tackle complex tasks, but their performance can vary significantly depending on the task at hand.”

Multi-LLM Medication Recommendations

ArXiv paper 2512.05066v1 explores collaborative approaches for clinical decision support. The research addresses the challenge that “individual large language models (LLMs) are susceptible to hallucinations and inconsi[stencies]” in healthcare applications, according to the abstract. The work focuses on improving reliability in AI-driven medication recommendations.

Psychological Analysis of Conversations

ArXiv paper 2512.04728v1 introduces a model for analyzing psychological elements in natural conversations. The research identifies a key challenge: “existing Vision-Language Models (VLMs) fail to resolve Articulatory-Affective Ambiguity, where visual patterns of speech mimic emotional expressions,” according to the paper’s abstract. The work aims to improve AI systems’ ability to distinguish between speech patterns and genuine emotional expressions.

All three papers represent cross-domain submissions to the arXiv AI category (cs.AI).