Given a strategically complex board game, human players can quickly learn to devise strategies after playing a few rounds. Autonomous agents require similar capabilities in realistic interactive environments, yet existing agent benchmarks often fail to fully capture such strategic and evolving decision-making scenarios. We present **PTCG-Bench**, a benchmark built on the Pok’emon Trading Card Game (PTCG) that evaluates LLM agents at two complementary levels: (1) their decision-making performance within a single complex environment, and (2) their ability to self-evolving through accumulated experience. We further include a modular harness ablation to better interpret agent performance without conflating it with model capability. Our experiments show that, although LLM agents can achieve non-trivial gameplay performance, sustained and stable self-evolution remains challenging, and performance is sensitive to harness design. We hope that PTCG-Bench will facilitate future research on harness-aware and self-evolving agents in realistic interactive environments.

Agentic RAG, which decomposes a complex question into dependency-aware sub-questions, has become the dominant paradigm for multi-hop QA. Yet existing systems can only repair errors at the failed step or simply return to the previous step and retry, and thus fail to handle two pervasive failure modes: inaccurate decomposition, where the plan is flawed before execution begins, and error propagation, where an upstream mistake surfaces downstream. We propose that addressing both requires global self-correction, the ability to identify the root cause of a failure and repair it at the right location. To this end, we propose **LoCoRAG**, a globally self-correcting agentic RAG framework. LoCoRAG maintains a persistent dual memory through a two-stage thinker-generator pipeline, separating what the corpus contains from what the system has concluded. Then, a memory-grounded failure localization module attributes each failure to specific causes and pinpoints its true origin. Together, LoCoRAG drives a self-correcting global replan that selects a matched repair, from local refinement to upstream rollback with downstream regeneration. Across existing multi-hop QA benchmarks, LoCoRAG achieves SOTA performance on all datasets and all metrics, with a 10.8 points average LLM-Acc gain over the strongest baseline on 3-hop-or-deeper scenarios.

“Large language models (LLMs) have become integral components of various AI solutions, with the reinforcement learning from human feedback (RLHF) stage playing a critical role in align-ing model outputs with human preferences. However, generating the human preference data required for RLHF is often costly and time-consuming due to its reliance on human evaluation.This study addresses this challenge within the dialogue scenarios of the fintech industry. We leverage rich, non-confidential, multi-turn dialogue data, such as call center dialogue records,which include associated business metrics (e.g., problem-solving rates, turnover ratios) to con-struct preference-aligned data. We introduce Self-Preference, an automated method for creating preference-aligned data guided by these objective business metrics. The approach involves clustering dialogue histories based on their semantic representations and calculating a well-designed conditional probability ratio that correlates sequences with business metrics to generate preference data. In contrast to traditional preference alignment data generation methods that depend on subjective human evaluations, Self-Preference significantly reduces labeling costs and mitigates model-induced biases. Experimental results indicate that models trained with Self-Preference generated data demonstrate a strong positive correlation with target business metrics, highlight-ing the method’s effectiveness in facilitating efficient, goal-oriented alignment of LLMs.”