arXiv 论文速递
分词器是语言模型的基础组件，但BPE与UnigramLM的优劣来源不明。本文沿两个正交维度分解分词质量，厘清目标函数与搜索算法各自的贡献，为分词器设计提供理论指导。
Two dominant tokenisation algorithms are used by modern language models: byte-pair encoding (BPE) and UnigramLM. These differ along two orthogonal axes: their optimisation objectiv…
· 2026-09-16
直接偏好对齐方法依赖梯度回传，计算与显存开销大。本文提出零阶优化范式实现LLM偏好对齐，绕开反向传播，为资源受限场景下对齐大模型提供高效新路径。
Direct preference alignment methods are widely used to align large language models (LLMs) with human preferences because of their computational and memory efficiency. However, like…
现有视频生成机器人方法只产生纯运动轨迹，缺乏接触力信息。本文融合视频与音频生成实现零样本力感知操作与数据生成，弥补生成式机器人学习在力觉模态上的关键空白。
Recent advances in video generation allow robots to learn manipulation trajectories from generated videos. However, these approaches produce purely kinematic trajectories that lack…
离线策略评估是强化学习落地的核心难题。本文证明即便日志数据充分覆盖隐藏状态，在历史依赖的日志策略下评估仍可能指数级困难，揭示了现有方法的根本局限。
Can a logged dataset visit every hidden state frequently and still be exponentially uninformative about a target policy's value? We show that it can when the logger depends on hist…
· 2026-09-16
Language agents remain brittle in interactive environments, where success requires long-horizon state tracking, valid action execution, and recovery from failed steps. We extend Sw…
· 2026-09-16
Computer-use agents increasingly operate software designed for people, but interfaces often leave actions or task state unclear to machine readers. We present Affora, a design syst…
Emergent coordinated behaviors of AI agents are starting to present critical safety risks. A key phenomenon driving these behaviors is the rapid formation and spread of beliefs abo…
Scaling laws predict how loss decreases with increases in computation. We show, contrary to conventional wisdom, that architectural interventions can modify scaling exponents in pr…
Factory work is a promising early scenario for embodied AI: assigning repetitive manual jobs to robots has clear economic payoff, and a structured station keeps the jobs tractable …
· 2026-09-16
奖励黑客随模型规模增长愈发隐蔽且后果严重，本文探究其是否在模型内部表征中留下可检测信号，为评估阶段实时监控与发现奖励黑客提供新方法，对LLM安全对齐至关重要。
As models scale, reward hacking becomes more frequent, more sophisticated, and more consequential. Does it leave a telltale signature in model representations? This work analyzes h…
Self-emulsifying drug delivery systems (SEDDS) can improve the oral bioavailability of poorly soluble drugs, but identifying high-performing formulations remains experimentally int…
· 2026-09-16
Artificial intelligence (AI) is increasingly integrated into healthcare systems worldwide, yet its successful clinical adoption depends critically on workforce readiness, particula…