Large Language Models | Xunyi's Academic Websitehttps://jiangxunyi.github.io/tag/large-language-models/Large Language ModelsHugo Blox Builder (https://hugoblox.com)en-usWed, 11 Nov 2026 00:00:00 +0000https://jiangxunyi.github.io/media/icon_hu12864870738162112581.pngLarge Language Modelshttps://jiangxunyi.github.io/tag/large-language-models/SymSkill: Skill Graph for Symbolic Music Understandinghttps://jiangxunyi.github.io/publication/symskill/Wed, 11 Nov 2026 00:00:00 +0000https://jiangxunyi.github.io/publication/symskill/<p>SymSkill organizes procedural music knowledge into a graph of skills to support language-model reasoning about symbolic music.</p>Toward Latent Language Model Skills Steering and Optimization: An Empirical Studyhttps://jiangxunyi.github.io/publication/skillvector/Thu, 03 Sep 2026 00:00:00 +0000https://jiangxunyi.github.io/publication/skillvector/<!-- <div class="alert alert-note"> <div> Create your slides in Markdown - click the <em>Slides</em> button to check out the example. </div> </div> --> <!-- Add the publication's **full text** or **supplementary notes** here. You can use rich formatting such as including [code, math, and images](https://docs.hugoblox.com/content/writing-markdown-latex/). -->Reasoning Pathologies in Large Language Models: A Diagnostic Perspectivehttps://jiangxunyi.github.io/publication/reasoningpathologies/Tue, 01 Sep 2026 00:00:00 +0000https://jiangxunyi.github.io/publication/reasoningpathologies/<p>A diagnostic study of reasoning failures in large language models.</p>The Geometry of Agent Skills: Non-Commutative Composition in Representation Spacehttps://jiangxunyi.github.io/publication/agentskillgeometry/Tue, 01 Sep 2026 00:00:00 +0000https://jiangxunyi.github.io/publication/agentskillgeometry/<p>A study of the geometry of agent skills and how their composition depends on order in representation space.</p>Can We Break LLMs Out of Self-Loops? Fine-Grained Reasoning Control with Activation Steeringhttps://jiangxunyi.github.io/publication/selfloops/Mon, 20 Jul 2026 00:00:00 +0000https://jiangxunyi.github.io/publication/selfloops/<p>SOPHIA detects when a model repeats unproductive reasoning states and intervenes through hidden-state steering. The method uses transitions between reasoning states to guide generation, improving answer quality and token efficiency.</p>RRPO: Reference-Relative Policy Optimization with Stratified Conditional Rolloutshttps://jiangxunyi.github.io/publication/rrpo/Mon, 20 Jul 2026 00:00:00 +0000https://jiangxunyi.github.io/publication/rrpo/<p>RRPO extends group-relative reinforcement learning to tasks without a direct correctness verifier. It compares candidate rollouts against positive and negative reference sets to construct training advantages for reasoning and open-ended generation.</p>Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learninghttps://jiangxunyi.github.io/publication/rolloutsurvey/Wed, 08 Apr 2026 00:00:00 +0000https://jiangxunyi.github.io/publication/rolloutsurvey/<p>This survey organizes rollout design for language-model reinforcement learning into generation, filtering, control, and replay. It connects design choices to reliability, coverage, and computational cost, with examples across reasoning and agent tasks.</p>When Benchmarks Age: Temporal Misalignment through Large Language Model Factuality Evaluationhttps://jiangxunyi.github.io/publication/bechage/Tue, 24 Mar 2026 00:00:00 +0000https://jiangxunyi.github.io/publication/bechage/<!-- <div class="alert alert-note"> <div> Create your slides in Markdown - click the <em>Slides</em> button to check out the example. </div> </div> --> <!-- Add the publication's **full text** or **supplementary notes** here. You can use rich formatting such as including [code, math, and images](https://docs.hugoblox.com/content/writing-markdown-latex/). -->