Tag: Long Horizon

Qiankai Xu, Qiguang Chen, Zixin Su, Wenhao Huang, Yue Gao, Jiaheng Liu · Aug 26, 2026 · Citations: 0

• The same reverse construction extends to instruction data and evaluation.
• Anchoring tasks in held-out papers yields PAW-Bench, an academic-writing benchmark whose tasks carry their own rubrics and checklists.

Jinpu Jiang, Xuan Wu, Wenhao Song, Bo Yang, You Zhou, Hongwei Ge · Aug 26, 2026 · Citations: 0

• To address this limitation, we propose ReliableRAG, which, to the best of our knowledge, is the first reliability-driven framework that mitigates deceptive misinformation in multi-hop QA through fine-grained evaluation of individual…

Can Xie, Yuyi Zhou, Wen Yang, Ziyi zhang, Siyao Song, Yingzhuo Deng · Aug 22, 2026 · Citations: 0

• Reinforcement learning with outcome-based objectives such as GRPO enables LLM-based agents to solve complex, long-horizon tasks, yet the reusable exploration patterns embedded in interaction trajectories are largely discarded after a single…

Xubin Chen, Yipeng Zhou, Wen Sun, Chengkai Huang, Xiaoming Fu, Quan Z. Sheng · Aug 21, 2026 · Citations: 0

• Experiments on benchmark datasets show that KREL consistently outperforms strong PLM-based and state-of-the-art LLM-based baselines.

Ji Liu, Puyuan Yang, Rongzhang Zheng, Fan Wang, Jinglin Wang, Muhammad A. Awad · Aug 21, 2026 · Citations: 0

• We present AsmEvo, an agentic assembly-level optimizer for AMD GPU kernels.
• Given an AMDGPU code object K0, AsmEvo reconstructs a reassemblable representation, proposes low-level edits with a long-horizon agent, rebuilds an ABI-preserving optimized object, and accepts candidates only after differential verification…

Lauren Pothuru · Aug 20, 2026 · Citations: 0

• Agentic retrieval-augmented generation (RAG) interleaves retrieval, reasoning, and answer generation across multiple hops.
• This paper introduces AgenticRAG-FP, an interventional benchmark for causal failure attribution in agentic RAG.

Brahim Touayouch, Mirette Moawad, Dmitry Akulov · Aug 20, 2026 · Citations: 0

• Abstract shows limited direct human-feedback or evaluation-protocol detail; use as adjacent methodological context.

Zhifa Geng, Subin Huang, Hao Guo, Junjie Chen, Sanmin Liu, Chao Kong · Aug 20, 2026 · Citations: 0

• Abstract shows limited direct human-feedback or evaluation-protocol detail; use as adjacent methodological context.

Seongjae Kang, Taehyung Yu, Sung Ju Hwang · Aug 20, 2026 · Citations: 0

• Customer-service LLM agents must follow organizational policy when acting on a user's behalf.
• Runtime safeguards can intervene on risky actions, but action-local checks do not guide an agent through a multi-step procedure.

Bo Qian, Yuting Wu, Shuang Zeng, Huaiyu Wan, Dalin Zhang, Jiqiang Liu · Aug 20, 2026 · Citations: 0

• Credit assignment is challenging in long-horizon agentic reinforcement learning, where supervision often comes only from final rewards.

Haonan He, Xinyue Fan · Aug 20, 2026 · Citations: 0

• For instance, LoRA-GA^2 surpasses the leading baseline by an average of 0.66 points on the GLUE benchmark, and outperforms the strongest baseline by 1.03 points on GSM8K and 0.87 points on HumanEval, respectively.

Haiyue Zhang · Aug 20, 2026 · Citations: 0

• Audited against causal ground truth from executed replay in a single-agent tool environment (ALFWorld), none of the step-level credit signals used to train LLM agents -- LLM-judge scores, outcome-conditioned logprob ratios, or the policy's…
• A confidence-only router recovers pivotal steps at chance level, but cuts judge cost by 13.1% per turn (14.0% per trajectory).

Zhuochun Li, Youngmin Ko, Ali Keramati, Nicola Ferri, Susana Palmaz Lopez Pelaez, Liang-Chun Tsai · Aug 20, 2026 · Citations: 0

• Recent agent benchmarks increasingly ground evaluation in executable environments, from code repair to web navigation, app APIs, and function calling.
• In this paper, we introduce Thinkingbox, a sandbox for tool-agent-user interaction that provides isolated MCP-compatible tool sessions, complete execution traces, and outcome evaluation over terminal backend state.

Xinyi Liu, Hooshang Nayyeri, Dilek Hakkani-Tur, Emine Yilmaz, Joo-Kyung Kim, Yifei Zhang · Aug 20, 2026 · Citations: 0

• Yet existing benchmarks typically evaluate prosodic perception, response appropriateness, and task-oriented dialogue in isolation, making it difficult to test whether prosodic evidence changes downstream decisions.
• We introduce Hear2Act, a unified evaluation protocol for text and spoken assistants with 480 persona-grounded scenarios, hidden user concerns, and objectively verifiable outcomes.

Bo Liu, Simon Yu, Yiding Jiang, Ao Qu, Andrew Zhao, Zichen Liu · Aug 19, 2026 · Citations: 0

• For language agents, existing training environment pools (hand-curated, statically synthesized, or frozen-verifier) keep the goal distribution fixed as the learner scales.
• We introduce SPADE (Self-Play in Adaptive Synthetic Executable Environments), a self-play RL framework in which a single LLM plays two roles: an Environment Designer that writes complete, long-horizon training environments as executable…

Zhu Zhang, Jixun Wang, Xiaoang Xu, Xiaorong Wang, Zihan Zhou, Zhiyuan Wang · Aug 19, 2026 · Citations: 0

• Across five long-context benchmarks, post-training with GC-OPD raises the five-benchmark averages of the official Qwen3-4B and Qwen3-8B checkpoints from 29.08 to 40.47 and from 35.12 to 44.65, respectively.

Mint-Agent Team, Kun Wang, Gavin Zhang, Yaze Geng, Lei Tang, Yaoyang Yi · Aug 17, 2026 · Citations: 0

• We present Mint-Agent, a family of finance-native agentic models designed around these two scales of financial intelligence.
• Across professional financial benchmarks, our models demonstrate two defining strengths: (1) Reliability: Mint-Ag achieves 98.33% on RFC-Bench, surpassing GPT-5.6-Sol and Claude-Opus-4.8 by 3.66 and 3.00 points; and (2) Executability:…

Yaxin Luo, Haobin Jiang, Jialv Zou, Xu Huang, Wenhao Yan, Haodong Li · Aug 13, 2026 · Citations: 0

• In this paper, we present AutoDesign, a framework that aligns with human design priors, where a meta-harness optimizer guides a code agent to recursively improve harness based on rollout feedback.
• Across seven controlled code-agent-model configurations, integrating the learned DesignHarness consistently improves performance, increasing the average PosterBench Score from 54.99 to 67.39 (+12.4%).

Yuto Nishida, Hirokazu Kiyomaru, Yusuke Oda, Takashi Kodama, Chaoran Liu, Daisuke Kawahara · Aug 13, 2026 · Citations: 0

• Abstract shows limited direct human-feedback or evaluation-protocol detail; use as adjacent methodological context.

Lei Bai, Jiaqi Cao, Chiyu Chen, Guanzhou Chen, Kai Chen, Guangran Cheng · Aug 13, 2026 · Citations: 0

• We present Intern-S2-Preview, a series of scientific agentic foundation models designed to support multimodal scientific understanding, reasoning, generation, and long-horizon tasks.
• Starting from the pretrained checkpoint, we apply a unified post-training pipeline consisting of supervised fine-tuning, scalable multi-task reinforcement learning (RL), black- and white-box agentic RL, and on-policy distillation.