Bing He

Minhua Lin, Juncheng Wu, Zijun Wang et al.

LLM agents are increasingly deployed as systems built around editable external harnesses, including prompts, skills, memories and tools, that shape task execution without changing model parameters. Harness self-evolution adapts such agents by updating these harnesses from execution evidence. Yet it remains unclear whether a model's base capability in task-solving predicts its capabilities in harness self-evolution: which models produce useful harness updates, and which actually benefit from them? We analyze two harness self-evolution capabilities: (i) harness-updating, the capability to produce useful persistent harness updates from execution evidence; (ii) harness-benefit, the capability to benefit from updated harnesses during task solving. Our analysis reveals two findings. First, harness-updating is flat in base capability: models from different capability tiers produce harness updates that lead to surprisingly similar gains; even Qwen3.5-9B's updates yield gains comparable to those of Claude Opus~4.6. Second, harness-benefit is non-monotonic in base capability: weak-tier models benefit little from updated harnesses, mid-tier models benefit most, and strong-tier models benefit less than mid-tier. We trace low gains at the weak tier to two failure modes: weak-tier models may fail to activate relevant harness artifacts, or activate them but fail to follow them faithfully. These findings suggest investing capability budget in the task-solving agent rather than the evolver, and targeting harness invocation and long-horizon instruction following in agent training. Our source code is publicly available at https://github.com/A-EVO-Lab/a-evolve/tree/release/harness-evolution.

Bing He, Sreyashi Nag, Limeng Cui et al.

E-commerce platforms typically store and structure product information and search data in a hierarchy. Efficiently categorizing user search queries into a similar hierarchical structure is paramount in enhancing user experience on e-commerce platforms as well as news curation and academic research. The significance of this task is amplified when dealing with sensitive query categorization or critical information dissemination, where inaccuracies can lead to considerable negative impacts. The inherent complexity of hierarchical query classification is compounded by two primary challenges: (1) the pronounced class imbalance that skews towards dominant categories, and (2) the inherent brevity and ambiguity of search queries that hinder accurate classification. To address these challenges, we introduce a novel framework that leverages hierarchical information through (i) enhanced representation learning that utilizes the contrastive loss to discern fine-grained instance relationships within the hierarchy, called ''instance hierarchy'', and (ii) a nuanced hierarchical classification loss that attends to the intrinsic label taxonomy, named ''label hierarchy''. Additionally, based on our observation that certain unlabeled queries share typographical similarities with labeled queries, we propose a neighborhood-aware sampling technique to intelligently select these unlabeled queries to boost the classification performance. Extensive experiments demonstrate that our proposed method is better than state-of-the-art (SOTA) on the proprietary Amazon dataset, and comparable to SOTA on the public datasets of Web of Science and RCV1-V2. These results underscore the efficacy of our proposed solution, and pave the path toward the next generation of hierarchy-aware query classification systems.

Long-Kai Huang, Rongyi Zhu, Bing He et al.

Protein Language Models (PLMs), pre-trained on extensive evolutionary data from natural proteins, have emerged as indispensable tools for protein design. While powerful, PLMs often struggle to produce proteins with precisely specified functionalities or properties due to inherent challenges in controlling their outputs. In this work, we investigate the potential of Activation Steering, a technique originally developed for controlling text generation in Large Language Models (LLMs), to direct PLMs toward generating protein sequences with targeted properties. We propose a simple yet effective method that employs activation editing to steer PLM outputs, and extend this approach to protein optimization through a novel editing site identification module. Through comprehensive experiments on lysozyme-like sequence generation and optimization, we demonstrate that our methods can be seamlessly integrated into both auto-encoding and autoregressive PLMs without requiring additional training. These results highlight a promising direction for precise protein engineering using foundation models.