Task knowledge is commonly stored either as text in the prompt or as an update to model weights. Text is modular but must be interpreted on every use, while weight adaptation makes the resulting capability difficult to load, remove, or share independently. We introduce KV-Skill, a design space of external factorized operators that a frozen language model reads through a lightweight interface. KV-Skill supports two complementary paths. Registration converts an authored text skill into a text-derived operator and trains a shared per-backbone interface. Reward learning develops a compact latent operator directly from task outcomes, with or without an authored skill. Neither path adds positions to the prompt. Across ten benchmarks and four backbones from three model families, converting text to a KV-Skill consistently makes the same procedural knowledge more effective. On Qwen3.5-4B LiveMath, registration reaches 77.2 accuracy, compared with 23.4 for the source text skill, 52.0 for SkillOpt, and 64.5 for SoftSkill. Under matched reward training and parameter budgets, KV-Skill gives the best result in seven of eight matched settings against soft prefixes, prefix tuning, and LoRA. A post-hoc rank analysis further shows that text-derived operators retain nearly all of their benefit with one task-aligned direction per injection layer, while matched random directions fail. Finally, one shared interface retains three independently loadable KV-Skills without measurable forgetting. These results show that task knowledge can be acquired from text or experience, compressed into an external operator, and deployed separately from the backbone. Code is available at: https://github.com/shawnzhg/KV-Skill
Zhaowei Han, Xiang Zhang, Bing Han et al.· 0 citations
Peered-reviewed literature provides reliable biomedical knowledge, which is essential for large language models (LLMs) in solving complex biomedical problems. However, how literature is retrieved and presented to LLMs can significantly influence their performance on domain-specific tasks.We present the Genomic Literature Knowledge Base (GLKB) and GLKB Agent to address these challenges. GLKB is a large-scale knowledge graph containing 14.6 million relationships among 3.2 million entities from 33 million PubMed abstracts and nine curated biomedical repositories. The current release includes articles published before March 2025. It supports diverse applications, including reinforcement learning, link prediction, and semantic embedding. The GLKB Agent is an agentic architecture that seamlessly connects LLMs to GLKB. It enables autonomous retrieval, reasoning, and deep research capabilities. Our evaluations demonstrate that the GLKB Agent dramatically improves LLM performance. Eight state-of-the-art LLMs achieve average accuracy gains of 27.5% on PubMedQA-HC, 24.8% on PubMedQA-Artificial, and 6.0% on BioASQ. Ablation tests confirm the agentic architecture is particularly effective for complex reasoning tasks. During datasource ablation tests, GLKB outperforms alternative data sources including PubMed, Wikipedia, and arXiv. Beyond question-answering, GLKB agent also demonstrates deep research capabilities through test-time reasoning. It generates comprehensive reports for literature reviews and hypothesis generation. The GLKB and GLKB Agent together provide a strong foundation for next-generation biomedical AI. Access is available at: https://glkb.org.
Yuanhao Huang, Zhaowei Han, Kevin Chang et al.· Proceedings of the 32nd ACM...· 0 citations