Learning GenAI via SOTA Papers

Title: SkillGLoW: Procedural-Family Skill Consolidation for Self-Improving Agents on Long-Horizon Task Streams Source: http://arxiv.org/abs/2609.02217v1 Summary:This research presents 'procedural-family skill consolidation' as a novel framework for self-improving agents. It addresses fundamental challenges in agent autonomy and generalization by enabling agents to learn, adapt, and consoli...

Title: Codebook Agent: Amortized Topology Design for LLM Multi-Agent SystemsSource: http://arxiv.org/abs/2609.02264v1 Summary: This paper introduces 'amortized topology design' as a novel architectural primitive for structuring interactions within LLM multi-agent systems. This approach promises significant efficiency and reasoning breakthroughs by optimizing how agents communicate and coor...

Title: Latent Recurrent Thoughts: Recurrent Refinement of Proposed Latents for Reasoning with Frozen LLMs Source: http://arxiv.org/abs/2609.01117v1Summary: This work proposes a novel reasoning loop that employs recurrent refinement of latent representations to enhance reasoning capabilities, even when utilizing frozen large language models. This represents a significant breakthrough in improving t...

Title: ARISE-RL: Agentic Rubric-Grounded Iterative Self-Evolution with Reinforcement LearningSource: http://arxiv.org/abs/2609.01058v1 Summary: This paper introduces a novel agentic reasoning framework that enables iterative self-evolution in AI agents using reinforcement learning and rubric-grounded evaluation. This provides a foundational mechanism for agents to autonomously learn, improve, and...

Title: Higher-Dimensional Rotary Position Embedding Source: http://arxiv.org/abs/2608.29715v1 Summary: This paper introduces an extension to Rotary Position Embedding (RoPE), a critical architectural primitive in Transformer models. This advancement can significantly enhance GenAI by improving how models process and understand complex, multi-dimensional data, leading to more capable and efficient...

Title: AutoCRAT: Within-trajectory Joint Control of Stochasticity and Compute for LLM ReasoningSource: http://arxiv.org/abs/2608.29988v1 Summary: This research proposes a novel framework for adaptively controlling stochasticity and computational resources during LLM reasoning. This represents a significant breakthrough in agentic reasoning, enabling more efficient, targeted, and adaptable decision...

Title: Toward Latent Language Model Skills Steering and Optimization: An Empirical Study Source: http://arxiv.org/abs/2608.29459v1Summary: This research explores methods for steering and optimizing latent language model skills, offering a deeper level of control over LLM capabilities. This represents a significant reasoning breakthrough for GenAI by enabling more efficient and targeted utilization...

Title: LiteSearch-VL: Small Multimodal Search Agents via Trajectory Distillation and Synthetic Step-DPO Source: http://arxiv.org/abs/2608.29357v1 Summary: This paper introduces novel techniques like trajectory distillation and synthetic Step-DPO to develop small, efficient multimodal search agents. Such advancements offer a significant efficiency and scaling breakthrough for Agentic AI, making sop...

Title: Omni-Interactive Universal EmbedderSource: http://arxiv.org/abs/2608.27044v1 Summary: This paper proposes a "universal embedder" capable of processing and unifying diverse, interactive data modalities into a singular representation. Such a primitive could fundamentally change how GenAI and Agentic AI systems perceive and represent the complex, multimodal world, enabling more versa...

Title: ASIL: Replacing Screenshot-and-Click with Structured State and Semantic Actions Source: http://arxiv.org/abs/2608.26991v1Summary: This work introduces a novel framework for agent-environment interaction, moving beyond brittle pixel-based methods to structured state and semantic actions. This paradigm shift is foundational for creating robust and intelligent AI agents capable of reliably per...

Title: ProgRouter: Online Progress-Guided Orchestration for Multi-Agent LLM Workflows under Quality-Cost Tradeoffs Source: http://arxiv.org/abs/2608.25992v1 Summary: This work proposes a novel framework for orchestrating complex multi-agent LLM workflows. By intelligently guiding agent interactions based on progress and optimizing for quality-cost tradeoffs, it provides a foundational mechanism fo...

Title: PolyMemDB: A Polyglot Database System for AI Memory Management Source: http://arxiv.org/abs/2608.25577v1 Summary: This paper introduces a polyglot database system specifically designed for comprehensive AI memory management, addressing a critical and pervasive bottleneck for scaling Generative AI and Agentic AI systems. It provides foundational infrastructure that enables agents to handle v...

Title: From General Agents to RCA Experts: A Self-Evolving Harness for Root Cause Analysis Source: http://arxiv.org/abs/2608.25661v1 Summary:This work proposes a "self-evolving harness" framework that enables general agents to specialize and become experts in specific domains, exemplified by Root Cause Analysis. This represents a novel agentic reasoning loop for continuous self-improveme...

Title: AsymSpec: Context-Asymmetric Speculative Decoding for Agentic LLMsSource: http://arxiv.org/abs/2608.26004v1 Summary: This paper introduces a novel speculative decoding technique optimized for agentic LLMs. It represents a significant efficiency breakthrough by intelligently managing context to speed up LLM inference, which is crucial for the iterative reasoning processes of AI agents.

Title: On-policy Distillation with Verifiable Reward Source: http://arxiv.org/abs/2608.24696v1Summary: This work presents a crucial advance in AI training by combining on-policy distillation with verifiable reward mechanisms. It offers a significant efficiency breakthrough for training generative models and agents while also addressing the foundational challenge of ensuring reliable and aligned AI...

Title: Joint Optimization of Tool Creation and Use for Large Language Model Agents Source: http://arxiv.org/abs/2608.24571v1 Summary:This paper introduces a foundational framework for LLM agents to not only utilize existing tools but also autonomously create new ones, then optimize both their creation and use. This represents a significant breakthrough in agentic reasoning, enabling more autonomou...

Title: TRACE: A Self-Evolving Skill Bank for Consistent, Limit-Aware LLM Agents Source: http://arxiv.org/abs/2608.22793v1Summary: This research introduces a novel agentic reasoning framework centered on a 'Self-Evolving Skill Bank,' which allows LLM agents to autonomously acquire, refine, and manage their capabilities. This directly addresses foundational challenges in Agentic AI by enabli...

Title: Object-Uni: A Unified Model for Object-Centric Spatial Understanding and Controllable Generation Source: http://arxiv.org/abs/2608.22757v1 Summary: This paper proposes a unified, object-centric model for spatial understanding and controllable generation, representing a new architectural primitive for GenAI. By enabling models to intrinsically reason about and manipulate individual objects,...

Title: Physical Agentic AI: An Architecture for Orchestrating a Robot Crew with LLMs Source: http://arxiv.org/abs/2608.22657v1 Summary: This paper proposes a novel architecture for "Physical Agentic AI," demonstrating how LLMs can orchestrate a crew of robots. It represents a foundational breakthrough by integrating advanced language models into physical systems, enabling complex multi-r...

Title: Coalition-Aware Skill Reliability for Self-Evolving Agents Source: http://arxiv.org/abs/2608.22610v1 Summary: This research introduces a foundational framework for "self-evolving agents" with "coalition-aware skill reliability." It presents a novel approach for agents to learn, adapt, and assess their skills dynamically in collaborative settings, paving the way for more...

Title: The Third Restructuring of Software Form: From the Three-Tier Architecture to Storage, Models, and Agents Source: http://arxiv.org/abs/2608.20201v1 Summary: This paper proposes a fundamental architectural paradigm shift for software systems, moving from traditional three-tier designs to a structure centered on storage, models, and agents. This redefines the primitives for building future so...

Title: PolicyGuide: From Guarding One Action to Guiding the Whole Workflow for Policy-Compliant LLM AgentsSource: http://arxiv.org/abs/2608.19861v1 Summary: This work introduces a novel agentic reasoning framework designed to guide LLM agents through entire workflows while ensuring continuous policy compliance. It represents a significant breakthrough in developing robust, reliable, and governable...

Title: Training-Free Inference-Time Self-Reflection and Cost-Bounded Early Stopping for Large Language Models Source: http://arxiv.org/abs/2608.18884v1Summary: This paper presents a training-free, inference-time self-reflection protocol that enhances LLM reasoning through a prompt-level generate-critique-revise cycle. It introduces a cost-bounded early stopping mechanism using a confirmation senti...

Title: SPADE: Self-Play in Adaptive Synthetic Executable Environments Source: http://arxiv.org/abs/2608.19197v1 Summary: This paper introduces a novel reinforcement learning framework for LLM agents where a single model plays dual roles as both environment designer and reasoning agent. By generating and interacting within its own synthetic executable environments, the framework provides a self-pla...

Title: Co-RL: Unsupervised Reasoning Emerges from Diverse Cohort in Multi-agent RL Source: http://arxiv.org/abs/2608.17253v2 Summary:This paper presents a breakthrough in how unsupervised reasoning capabilities can emerge naturally within multi-agent reinforcement learning systems. It introduces a novel framework where diverse cohorts of agents learn to reason without explicit supervision, represe...