
A radiotherapy head mask on the couch, red alignment lasers crossing it · Photo: National Cancer Institute
2026 年一項 Nature 研究發現,Drosophila erecta 果蠅的雄蠅會被食物氣味引發性興奮,群體動態再加以放大;少了這樣的環境,就不會交配。


Rainbow light swirl spinning on black · Photo: Flyd
Stream-R1 依獎勵模型的評分,為每段訓練影片、每個影格和區域加權,在相同速度下於基準測試小幅勝過直接對照的基準方法。
arXiv 預印本 · 2026年5月5日

Photo: Pascal Bernardon
發表在 Physical Review Letters 的計算發現,偏心旋近在最內側軌道附近出現共振,會加強純量雲在重力波上留下的痕跡。
Physical Review Letters · 2026年10月20日

Photo: Arturo Mendez
一篇預印本建構出「違規捷徑可證明報酬更高」的代理任務,發現只要對任何有紀錄的違規施以簡單懲罰,就能大幅減少這類違規。
arXiv 預印本 · 2026年11月9日

Gloved hands hold up a six-well plate of purple-stained cell cultures with plaques · Photo: Cdc
2026 年發表於 Nature 的研究發現,在培養的人類細胞中逐一干擾 18 個自閉症基因,會以相似方式改變基因活性,並認為 X 染色體上的 ZFX 基因可能具保護作用。
Nature · 2026年11月4日

Photo: Matthew Stephenson
一項 2026 年的研究計算四階重力理論的重力波輻射:含鬼場的版本需要質量高於約 10^-11 eV 的額外模態,無鬼場版本的耦合常數則受到上限約束。
Physical Review Letters · 2026年11月4日套用精簡訓練配方後,一個現有推理模型在 IMO 2025 和 USAMO 2026 的題目上拿到 42 分中的 35 分,但仰賴冗長且昂貴的起草與檢查循環。
A central challenge in adapting large language models (LLMs) is capability preservation: an adaptation can achieve its target while degrading general capabilities. We introduce CurvDelta, a scalable framework that incorporates the preservation constraint into the parameterization of the model update. By analyzing the local geometry of…
System One models such as Jev return decisions through fast, parallel prediction. In this paper, we introduce Jeveration, showing that the same decision paradigm can also generate high-quality images in a single forward pass. Jeveration builds on a frozen vision-language model, repurposing its visual-semantic knowledge for holistic image…
Guowei Yang, Ruihan Chen et al.
Anna Fedotova, Alexey Brazhe et al. · Jiaxing University
Rubén López-Sandoval, Marius F. Harter et al.
Taylor J. Hoyt, David Rubin et al. · Lawrence Berkeley National Laboratory
NPU design-space exploration for LLM inference mainly relies on an expensive simulator that often returns only scalar feedback such as latency and power. Such feedback can rank evaluated configurations, but gives little information for deciding why a configuration is slow or which architecture parameter should be changed next. This…
Elizabeth Mieczkowski, Katherine M. Collins et al. · Massachusetts Institute of Technology
Rahul Gupta, Judith L. Racusin et al. · Brera Astronomical Observatory
Large language model (LLM) agents routinely interact with untrusted external data, exposing them to indirect prompt injection (IPI) attacks that hijack agents into executing attacker-chosen tasks. To uncover these vulnerabilities at scale, automated red-teaming methods often leverage reinforcement learning (RL) to generate attacks that…
Fares Bougourzi, Abdenour Hadid · Sorbonne University Abu Dhabi
Paul L. Ebert, Yasir Iqbal et al.
Marta Perez‐Alcantara, Sam J. Washer et al. · Open Targets
We present Xing 4, a series of large language models that extends the TeleChat lineage through coordinated advances in model architecture and training infrastructure on Ascend. This report describes the design principles and systems engineering behind the series, spanning architecture selection, large-scale pretraining, and reasoning and…
Maximilian Weimar, Huan‐Li Zhou et al. · TU Wien
Miao Yu, Maziar Heidari et al. · Johannes Gutenberg University Mainz