Agentic Policy Optimization via Instruction-Policy Co-Evolution

Researchers have found that large language models (LLMs) make decisions about networking and friendship in ways that closely resemble human behavior, both in synthetic simulations and real-world contexts. This suggests that LLMs can replicate social decision-making processes similar to those of people.

Recent studies reveal that while AI, particularly generative AI, has improved in accuracy, its flawed reasoning processes pose significant risks in critical sectors such as healthcare, law, and education. These findings highlight the need for a deeper understanding of AI's decision-making mechanisms.

A new framework named CORE has been proposed for trajectory representation learning (TRL), which aims to enhance the encoding of raw trajectory data into low-dimensional embeddings by integrating context-aware route choice semantics. This approach addresses the limitations of existing TRL methods that treat trajectories as static sequences, thereby enriching the semantic representation of urban mobility patterns.

A comprehensive review on missing data imputation highlights the challenges posed by incomplete datasets across various fields, including healthcare and e-commerce. The study synthesizes decades of research, categorizing imputation methods from classical techniques to modern machine learning approaches, emphasizing the need for a unified framework to address missingness mechanisms and imputation goals.

A recent study has introduced influence functions as a method for efficient data selection in reasoning tasks, particularly for fine-tuning large language models (LLMs) on chain-of-thought (CoT) data. This approach aims to define data quality more effectively, moving beyond traditional heuristics like problem difficulty and trace length. Influence-based pruning has shown to outperform existing methods in math reasoning tasks.

Recent research has formalized the role of synthetically-generated data in training large language models (LLMs), highlighting that without proper curation, model performance can plateau or collapse. The study introduces a theoretical framework to determine the necessary curation levels to ensure continuous improvement in LLM performance, drawing inspiration from the boosting technique in machine learning.

A recent study published in Nature — Machine Learning found that human researchers outperformed large language models in writing a medical systematic review during a comparative multitask assessment. This research highlights the limitations of current AI capabilities in complex academic writing tasks, particularly in the medical field.