Adversarial Speculative Reasoning Improves Long Horizon Agents
Subconscious Systems Technologies, Inc.(潜意识系统科技公司);Massachusetts Institute of Technology(麻省理工学院);Princeton University(普林斯顿大学);Northeastern University(东北大学)
Hongyin Luo, Wei Fang, Lijie Yang, Owen Stepan, Daniel Suit, Jeff Krapf, Dana Wensberg, Jack O'Brien
Speculative decoding has been treated purely as a systems optimization that speeds up large language model decoding without changing the model behavior. Under this view, the growing effort to train better draft models has been focused solely on acceptance rate. We argue that this view does not hold true for long-horizon agent reasoning. Measuring frontier open-weight models on complex agentic tasks with different speculative decoding approaches, we found that numerical noise grows within verification windows, and the draft acceptance results of the thinking process can change the output distribution of the model after 30 steps, thus noticeably influencing the reasoning accuracy of long-horizon agents in random directions. In this work, we improve the reasoning capability of target models by controlling such noise. While we cannot change the overall token distribution verified by the target model, we can encode critical reasoning with highest precision by training the draft model to be rejected. We identify these points in agent traces and train the draft model to favor adversarial preference. At inference the trained draft and the target part ways precisely at these decision points, verification restarts, so that critical thinking tokens are sampled from the target model and encoded under its least noisy conditions. This approach makes critical reasoning better indexed and more durable over a long horizon. On complex agentic coding benchmarks, our approach raises end-to-end success rate with no change to the target model or its output distribution. Our results recast the speculative decoding mechanism from a pure accelerator into an active allocator for context precision that improves the accuracy of long-horizon agents by training 0.06% to 0.28% of hosted parameters.
HeroBench: A Benchmark for Long-Horizon Planning and Structured Reasoning in Virtual Worlds
Artyom Sorokin(AXXX);Lomonosov Moscow State University(罗蒙诺索夫莫斯科国立大学);Higher School of Economics(高等经济学院);Kurchatov Institute(库尔斯克研究所);Independent Researcher(独立研究者)
Speculative Rollback Correction for Quality-Diverse Web Agent Imitation
Beihang University(北京航空航天大学);Institute of Software, Chinese Academy of Sciences(中国科学院软件研究所);The Hong Kong University of Science and Technology(香港科技大学);Northwestern Polytechnical University(西北工业大学);Tsinghua University(清华大学);The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州));Peking University(北京大学)
Off-Policy Actor-Critic for Adversarial Observation Robustness: Virtual Alternative Training via Symmetric Policy Evaluation
Department of Information Science, Kyoto University, Kyoto, Japan(京都大学信息科学系);ATR Neural Information Processing Laboratories, Kyoto, Japan(ATR神经信息处理实验室);International Research Center for Neurointelligence, The University of Tokyo, Tokyo, Japan(神经智能国际研究中心)
Evo-Attacker: Memory-Augmented Reinforcement Learning for Long-Horizon Tool Attacks on LLM-MAS
Beihang University(北航);Beijing University of Posts and Telecommunications(北京邮电大学);China Academy of Information and Communications Technology(信息通信技术研究院)