Discovering improved policy optimization algorithms for language models remains a costly manual process requiring repeated mechanism-level modification and validation. Unlike simple combinatorial code search, this problem requires searching over algorithmic mechanisms tightly coupled with training dynamics while reusing empirical evidence across iterations. We propose POISE a closed-loop framework for automated discovery of policy optimization algorithms for language models. POISE maintains an archive that links each candidate design with its implementation, evaluation evidence, and analysis summary, enabling later iterations to build on both successful and failed designs. In mathematical reasoning experiments starting from GRPO, POISE evaluates 64 candidate algorithms and discovers improved mechanisms, including analytic-variance scaling and validity masking. The best variant improves weighted Overall from 47.8 to 52.5 (+4.6) and increases AIME25 pass@32 from 26.7% to 43.3%, demonstrating the feasibility of automated policy optimization discovery while supporting interpretable design principles.

LLM role-playing, i.e., using large language models (LLMs) to simulate specific personas, has emerged as a key capability in various applications, such as companionship, content creation, and digital games. While current models effectively capture character tones and knowledge, simulating the inner thoughts behind their behaviors remains a non-trivial challenge. Towards cognitive simulation in LLM role-play, previous efforts have mainly suffered from two critical deficiencies: the lack of high-quality datasets with explicit reasoning traces and the absence of reliable reward signals aligned with human preferences. In this paper, we propose HER (Human Emulation Reasoning), a unified framework for cognitive-level persona simulation. HER introduces a dual-layer thinking mechanism that strictly distinguishes characters’ first-person thinking processes from LLMs’ third-person reasoning. To bridge the aforementioned gaps, we curate a reasoning-augmented role-playing dataset via a reverse engineering strategy for supervised learning, and construct human-aligned evaluation principles and preference-based reward models for role-play reinforcement learning. Leveraging these resources, we train HER models based on the Qwen3-32B backbone via a hybrid paradigm of supervised learning (SL) and reinforcement learning from human feedback (RLHF). Extensive experiments validate the effectiveness of our approach. Notably, our models significantly outperform the Qwen3-32B baseline, achieving a 30.26% on the CoSER benchmark and a 14.97% on the MiniMax Benchmark. Our datasets, evaluation principles, and trained models will be released to facilitate future research in cognitive-level LLM role-playing.

Behavioral problems in deployed LLMs are often discovered when only limited supervision is available and repeating full alignment training is impractical. Existing post-hoc methods have complementary limitations: activation steering lacks input adaptivity, while standard LoRA learns without an explicit behavioral direction prior and can be unstable in low-data settings. We introduce PR-LoRA, a direction-regularized LoRA variant that initializes LoRA B matrices with per-layer contrastive behavioral directions and anchors them with an L₂ penalty, while leaving A freely trainable. We evaluate PR-LoRA on seven CAA behaviors across Qwen3 and LLaMA models against activation steering and parameter-efficient adaptation baselines. On Qwen3-8B, PR-LoRA lowers the seven-behavior macro-average undesirable-behavior score from .405 to .374 relative to standard LoRA, a 7.8% relative reduction, and outperforms DPO-LoRA and LoReFT under the tested low-update budget.

Maintaining persona consistency across multi-turn dialogues remains a core challenge for role-playing language models. Off-policy distillation from external teachers incurs distribution mismatch that compounds across dialogue turns, while reinforcement learning struggles with reward ambiguity inherent in subjective persona fidelity. We propose OSPD, an on-policy self-distillation framework where the same model serves as both teacher and student under asymmetric information: the teacher receives a complete character profile while the student sees only a brief summary, and the student generates trajectories from its own policy. We find that teacher confidence in role-playing dialogue exhibits a bimodal structure—sharply peaked at character-critical tokens yet diffuse at generic utterances—and introduce role-aware divergence switching to match this structure. A progressive trait masking curriculum further forces staged internalization of character knowledge along semantic dimensions. Experiments on CharacterBench, CharacterEval, and SocialBench show that OSPD substantially improves persona consistency over supervised fine-tuning and multi-turn RL baselines, without requiring any external teacher or reward model.

Recent advances in Large Language Models (LLMs) have demonstrated strong capabilities in tasks such as code generation and mathematical reasoning. However, their potential to internalize structured spatial knowledge remains underexplored. This study investigates whether LLMs, grounded in locally relative human observations, can construct coherent global spatial cognition by integrating fragmented relational descriptions. We focus on two core aspects of spatial cognition: spatial perception, where models infer consistent global layouts from local positional relationships, and spatial navigation, where models learn road connectivity from trajectory data and plan optimal paths between unconnected locations. Experiments conducted in a simulated urban environment demonstrate that LLMs not only generalize to unseen spatial relationships between points of interest (POIs) but also exhibit latent representations aligned with real-world spatial distributions. Furthermore, LLMs can learn road connectivity from trajectory descriptions, enabling accurate path planning and dynamic spatial awareness during navigation.

Language agents powered by large language models (LLMs) are increasingly valuable as decision-making tools in domains such as gaming and programming. However, these agents often face challenges in achieving high-level goals without detailed instructions and in adapting to environments where feedback is delayed. In this paper, we present SELFGOAL, a novel automatic approach designed to enhance agents’ capabilities to achieve high-level goals with limited human prior and environmental feedback. The core concept of SELFGOAL involves adaptively breaking down a high-level goal into a tree structure of more practical subgoals during the interaction with environments while identifying the most useful subgoals and progressively updating this structure. Experimental results demonstrate that SELFGOAL significantly enhances the performance of language agents across various tasks, including competitive, cooperative, and deferred feedback environments.

To advance personalized applications such as recommendation systems and user behavior prediction, recent research increasingly adopts large language models (LLMs) for human-readable persona modeling. In dynamic real-world scenarios, effective persona modeling necessitates leveraging streaming behavior data to continually optimize user personas.However, existing methods—whether regenerating personas or incrementally extending them with new behaviors—often fail to achieve sustained improvements in persona quality or future behavior prediction accuracy. To address this, we propose DEEPER, a novel approach for dynamic persona modeling that enables continual persona optimization. Specifically, we enhance the model’s direction-search capability through an iterative reinforcement learning framework, allowing it to automatically identify effective update directions and optimize personas using discrepancies between user behaviors and model predictions.Extensive experiments on dynamic persona modeling involving 4,800 users across 10 domains highlight ’s superior persona optimization capabilities, delivering an impressive 32.2% average reduction in user behavior prediction error over four update rounds—outperforming the best baseline by a remarkable 22.92%.