AI·Coverage

Research/paper/2026-09-16

Apple ML Research publishes papers on LLM agents, memory, behavior, and ML methods

Apple Machine Learning Research has recently released five new papers, showcasing a range of advancements in artificial intelligence. The publications include research on improving agentic LLM systems with shared selective persistent memory and introducing Glyph, a multi-strategy agent for enterprise data cataloging. Further studies explore how value induction reshapes LLM behavior and present DACA-GRPO, a method for reinforcement learning in diffusion language models. Another paper details "Trajectory as the Teacher," a novel technique for few-step discrete flow matching. These first-party research outputs collectively demonstrate Apple's ongoing contributions to diverse areas of machine learning.

What every outlet reports

  • Apple Machine Learning Research published five new papers
  • Research covers LLM agents, memory, and behavior
  • New methods for reinforcement learning in diffusion models were introduced
  • A novel technique for few-step discrete flow matching was presented
  • An agentic system for enterprise data cataloging was detailed

All 5 articles

pragmatist5
Shared Selective Persistent Memory for Agentic LLM SystemsoriginalApple Machine Learning Research · research
How Value Induction Reshapes LLM BehaviouroriginalApple Machine Learning Research · research

Related stories