The field of large language model reasoning is moving towards more innovative and interactive approaches. Researchers are exploring ways to enhance the diversity and coherency of reasoning processes, particularly in areas such as mathematical reasoning and role-playing. One notable direction is the development of dialogue-based reasoning models, which aim to improve interpretability and facilitate more intuitive human interaction. Another area of focus is the representation of character logic as structured, executable functions for behavioral decision-making, offering advantages such as persistence, updatability, and controllable randomness. Inspired by human learning strategies, novel curriculum learning and reinforcement learning approaches are being proposed to advance large language model reasoning capabilities. Noteworthy papers include:

• DialogueReason, which proposes a dialogue-based reasoning paradigm to boost diversity and coherency of the reasoning process.
• Codifying Character Logic in Role-Playing, which introduces codified profiles for role-playing, offering significant benefits in improving persistence, updatability, and behavioral diversity.
• Learning Like Humans, which proposes adaptive difficulty curriculum learning and expert-guided self-reformulation to enhance large language model reasoning capabilities.