Weinan Zhang

Semantic-ID generative recommenders represent each item as a short sequence of discrete semantic tokens and predict the next item by autoregressively generating this token sequence. This paradigm enables a unified generation interface for item IDs, histories, and item text, but it also creates a structured optimization bottleneck during reward-based post-training: when an early semantic token enters the wrong branch of the item-token space, finite rollout groups rarely reach the ground-truth item, so group-relative optimization receives identical zero rewards and produces no useful advantage. We propose Hint-Conditioned Generative Recommendation (\method), a semantic-ID generative recommendation framework that recovers learning signal for such hard training instances. HCGRec diagnoses each instance with checkpoint rollouts and supplies a minimal target-prefix hint only when the current generator cannot reach the correct item. The model then generates the unhinted suffix under the hinted semantic branch, turning zero-reward groups into informative comparisons over item-token completions. Hinting also changes token identity: hinted prefix tokens are oracle-provided item context, while unhinted suffix tokens are sampled generation actions. We therefore introduce hint-aware credit decomposition, using supervised learning to preserve item-semantic and prefix-structure alignment for hinted tokens and GRPO to optimize the sampled suffix. Experiments on sequential recommendation benchmarks show that HCGRec substantially improves over supervised fine-tuning and vanilla reward-based post-training, while reducing zero-advantage training samples from over 70% to below 20%. The code is accessible at https://github.com/WncFht/GRec.

Generative recommendation has become a promising paradigm for producing complete recommendation lists, yet providing proper feedback to multiple candidates remains challenging. A key difficulty is the conflict between matching users' historical preferences and maintaining content diversity. Existing reward modeling and re-ranking methods usually rely on fixed objective aggregation or single-model judgment, which limits their flexibility and explainability in resolving such conflicts. To address this issue, we propose Debate4Rec, a multi-agent debate framework for feedback and candidate-list selection in generative recommendation. Debate4Rec acts as an external list-level feedback signal producer that provides interpretable feedback and selects the best candidate list. It decomposes list evaluation into two specialized agents: a Preference Agent for assessing alignment with user interests, and a Diversity Agent for evaluating category coverage, redundancy, and list richness. The agents first make independent judgments and then conduct a structured two-round debate to exchange evidence, revise decisions, and reach a final selection. Experiments on ML-1M and a Kuaishou short-video dataset show that Debate4Rec achieves competitive and often better performance than the LLM-based and traditional re-ranking baselines under the candidate-list selection setting. Ablation studies further confirm the effectiveness of role-specialized agents, structured debate, and the feedback-selector mechanism.

Embodied foundation models now allow robots to act on open-ended instructions, yet success on long-horizon physical tasks also depends on the runtime around the model, which organizes the task, invokes skills, guards execution, and retains experience. Digital agent research treats this runtime, known as the harness, as a first-class design object that can be engineered and even improved from experience, since the same model can perform very differently under different harnesses. Such harness designs do not carry over directly, because physical execution is embodiment-specific, partially observed, time-constrained, and often irreversible, which also makes experience costly to learn from. Embodied AI systems increasingly build similar runtimes, yet the field still lacks a unified view of them. This survey treats the physical agentic harness as a research object in its own right. We define it as the runtime system that maintains task and execution state for an embodied foundation model and mediates the model’s interaction with the physical world through one or more physical bodies. We then review over 200 works and organize them by six functional responsibilities: task orchestration, digital resources, embodied capabilities, execution governance, digital-physical interaction, and persistent evolution. This organization offers a common lens for comparing methods across communities, making explicit how model decisions reach physical execution and how grounded evidence flows back. The reviewed benchmarks and applications provide limited evidence of individual runtime contributions, particularly the effects of execution governance and sustained gains from persistent evolution. We close with open problems in model-harness co-design, capability composition, real-time decision making, distributed execution, and continual improvement from physical experience.

With the emergence of deep learning, semantic code search has been widely adopted in both academia and industry. These approaches embed natural-language queries and code snippets into a shared embedding space and retrieve results based on vector similarity. Despite their strong performance on benchmark datasets, they often suffer from poor explainability and generalization. Retrieved code may appear semantically similar yet miss critical functional requirements of the query, while providing no explanation of why the result was retrieved. Moreover, such failures become more severe under distribution shift, where models struggle to generalize to unseen benchmarks. In this work, we propose XSearch, an intrinsically explainable code search framework. Our key insight is that, by relying on global embedding similarity, all existing retrievers inherently take an inductive view. They learn statistical patterns, rather than truly understand the query's functional requirements. Therefore, we address the problem by reformulating code search as a deductive concept alignment problem. At a high level, XSearch (i) identifies functional concepts in the query and (ii) explicitly aligns them with corresponding code statements. This explain-then-predict design not only produces inherent concept-level explanations, but also mitigates shortcut learning that harms out-of-distribution generalization. We train an encoder with explicit concept-alignment objectives and perform retrieval through explicit matching between query concepts and code statements. Experiments show that, when trained on CodeSearchNet with a small model size (GraphCodeBERT with 125M parameters), XSearch improves performance on out-of-distribution benchmarks from 0.02 to 0.33 (15×) over eight state-of-the-art retrievers, and consistently outperforms both encoder-based and decoder-based baselines with up to 7B parameters. A controlled user study further demonstrates that concept-alignment explanations enable users to accept or reject retrieved results both faster and more accurately. The source code is publicly available at https://github.com/code-philia/Xsearch.

Issue localization, which identifies code locations requiring modification from issue descriptions, is a critical step in automated software maintenance. Existing approaches predominantly attempt to directly align issue descriptions with code elements, yet often struggle due to the inherent abstraction gap between the issue description and code implementation. Seeking alternative signals, our theoretical analysis suggests that test suites can serve as executable proxies for requirements, reducing localization uncertainty by 7.73 bits of entropy on average. A large-scale empirical study on 18 repositories validates this premise: existing tests cover 96.98% of ground-truth files, and the two-hop pathway yields stronger semantic connectivity than direct matching in 82.4% of cases. Despite their potential, leveraging tests for localization faces two key challenges: the semantic gap separating issue descriptions from test identifiers, and the substantial noise in execution traces from infrastructure code. To address these, we propose IssueExec, which bridges the semantic gap through domain-knowledge-enhanced test representations and filters noise via hierarchical trace analysis. Experiments on SWE-bench Lite show that IssueExec achieves state-of-the-art performance, improving function-level Recall@1 by 41.57% over the strongest baseline. When integrated into the Agentless pipeline, IssueExec resolves 17.72% more issues, demonstrating practical downstream benefits.