Personalizing large language models (LLMs) requires reflecting stable user preferences while adapting their expression to each target context. Soft-prompt personalization has become a practical route to this goal, but existing methods produce soft prompts in which stable preferences, target-induced content, and context-dependent expression are bundled into an entangled conditioning signal without separately addressable components, leaving inference with no principled space to intervene on individual aspects of personalization. We propose Factorized soft prompts with Latent Control (FaLCon), an LLM personalization framework built on two coupled stages. First, factorized soft prompts replace this entangled conditioning with a set of separately addressable preference factors, forming a structured control interface. Second, inference-time latent control treats the resulting factor weights as a continuous action space and adapts them per instance from teacher-forced likelihood feedback, keeping the LLM frozen and updating only a low-dimensional factor-weight posterior without back-propagating through the LLM. Experiments on multiple personalized generation tasks reveal that FaLCon consistently outperforms strong baselines, validating the individual necessity and joint effectiveness of the structured interface and the inference-time controller.

Large Language Models (LLMs) have recently shown strong reasoning capabilities, motivating their use in complex decision-making environments. StarCraft II (SC2), with its massive state-action space and partial observability, is a challenging testbed. However, existing LLM-based SC2 agents primarily focus on improving the policy itself, leaving the integration of a learnable, action-conditioned dynamics model into the decision loop largely unexplored. In this work, we propose StarWM and StarWM-Agent, and conduct the first systematic study of the learnability and decision utility of player-view, action-conditioned textual world models for SC2. StarWM predicts short-horizon future observations under partial observability. StarWM-Agent integrates StarWM into a lightweight Generate-Simulate-Refine loop for foresight-driven policy refinement. Extensive experiments show that StarWM substantially outperforms zero-shot baselines across multiple dimensions, while StarWM-Agent achieves consistent win-rate gains of 30%, 15%, and 30% against the SC2 built-in AI at Hard (LV5), Harder (LV6), and VeryHard (LV7), respectively. Additional analyses show that our method is complementary to existing history-summarization SC2-agent approaches, with their combination achieving the strongest online performance in our evaluation. Code is available at https://github.com/yxzzhang/StarWM.

Providing a reliable explanation for clinical diagnosis based on the Electronic Medical Record (EMR) is fundamental to the application of Artificial Intelligence in the medical field. Current methods mostly treat the EMR as a text sequence and provide explanations based on a precise medical knowledge base, which is disease-specific and difficult to obtain for experts in reality. Therefore, we propose a counterfactual multi-granularity graph supporting facts extraction (CMGE) method to extract supporting facts from irregular EMR itself without external knowledge bases in this paper. Specifically, we first structure the sequence of EMR into a hierarchical graph network and then obtain the causal relationship between multi-granularity features and diagnosis results through counterfactual intervention on the graph. Features having the strongest causal connection with the results provide interpretive support for the diagnosis. Experimental results on real Chinese EMR of the lymphedema demonstrate that our method can diagnose four types of EMR correctly, and can provide accurate supporting facts for the results. More importantly, the results on different diseases demonstrate the robustness of our approach, which represents the potential application in the medical field.