arXiv:2608.09696v1 Announce Type: new Abstract: Predicting the answer to interventional “what if” questions — the outcome of an action never taken —…
Category: cs.AI updates on arXiv.org
Second-Order Muon Done Right: A Principled Marriage of Spectral Geometry and Curvature
arXiv:2608.09763v1 Announce Type: new Abstract: Muon’s polar update is exact for an unweighted spectral geometry. We introduce GO-MUON, which uses a…
Avalon-ToM-Bench: Evaluating Fine-Grained Theory of Mind via Asymmetric Game Mechanics
arXiv:2608.09638v1 Announce Type: new Abstract: Theory of Mind (ToM) is essential for agent interactions, yet existing evaluations either rely on static…
Hallucination-Free GUI Grounding via Regression-Free Layout-Aware Matching
arXiv:2608.09654v1 Announce Type: new Abstract: GUI agents are shifting from metadata-dependent large language models to purely visual multimodal large…
Open Evaluation Agent: Efficient and Promptable Evaluation of Visual Generative Models
arXiv:2608.09666v1 Announce Type: new Abstract: Recent advances in visual generative models have enabled high-quality image and video generation, but…
Rethinking Self-Evolving Agents: Do We Still Need Prescribed Optimization Pipelines?
arXiv:2608.09629v1 Announce Type: new Abstract: Self-evolving agents are usually built around prescribed optimization pipelines: the framework decides how…
Adaptive Semantic Capacity Allocation for Parallel Generative Recommendation
arXiv:2608.09685v1 Announce Type: new Abstract: Autoregressive semantic ID recommenders are constrained by expensive beam-search decoding, which limits…
From Sweep to Seam: Interleaved Cross-Block Post-Training Quantization
arXiv:2608.09595v1 Announce Type: new Abstract: Compressing large language models to two bits or fewer is increasingly feasible through block-wise…
ElasticBack: Stealthy Conditional Backdoor in LLM-Agent Skills via Coupled Trigger-Rule Optimization
arXiv:2608.09577v1 Announce Type: new Abstract: Agent skills, bundles of instructions and resources that an LLM agent loads on demand, form an emerging…
CoRCi: Cross-Reconstruction of Coherent Interests Modeling in Cross-Domain Sequential Recommendation
arXiv:2608.09580v1 Announce Type: new Abstract: Cross-Domain Sequential Recommendation (CDSR) aims to alleviate data sparsity by transferring dynamic user…