Prépublication
· 2026 arXiv
EN
Systems that combine several large language models (LLMs) often use one context for many questions. When the context is passed between models as text, the receiving model processes all of its tokens again for every question. Avoiding this cost requires a conte...
Huawei Lin, Jinting Liu et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
LLM agents that compute the right answer can still answer wrongly. Agent harnesses increasingly keep long tool outputs out of the context window, saving them to files that the agent searches or reads on demand. We ask where this costs accuracy with a real scie...
Ke Zhang, Sahchit Chundur et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
We propose KronLSTM, a linear attention architecture based on the Kronecker Delta Rule, a novel update rule for associative matrix memories. KronLSTM extends the xLSTM family with a memory-revision mechanism that corrects targeted associations while modifying...
Anamaria-Roberta Hartl, Levente Zólyomi et al.
#cs.LG
Prépublication
· 2026 arXiv
EN
Many modern language models have adopted hybrid architectures that incorporate both linear-time recurrence and softmax attention. Modern recurrence layers provide a highly expressive mixing mechanism while ensuring high throughput by compressing contextual inf...
Xingyu Shen, Yingfa Chen et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Frontier agents retrieve structural metadata well and compute geometry from coordinates less reliably. We ask how much of a benchmark score reflects that gap and how much reflects the evaluation configuration around the model. We study ProteinGeoBench, a suite...
Matt Turk, Brandon L. Neel et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
Agent learning is shifting from learning from human data toward learning through interaction with the world. However, scaling such learning directly in real-world computer systems is challenging and can introduce substantial security risks. We therefore propos...
Zhaoyang Wang, Qianhui Wu et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
Context compression improves long-context efficiency but may introduce faithfulness issues through an information bottleneck. Existing evaluations rely on QA metrics, but a reader can still answer correctly from an unfaithful compression, so these metrics unde...
Ziqing Man, Di Wu et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Cosmic filaments connect the nodes of the large-scale web and are thought to harbour a substantial fraction of the low-z baryon budget, yet a physically motivated framework for organising their diverse gas properties has been lacking. Here we show that the lin...
Daniela Galárraga-Espinosa, Daisuke Nagai et al.
#astro-ph.CO
Prépublication
· 2026 arXiv
EN
We investigate the impact of turbulence and magnetic field on the formation of unipolar outflows in the gravitational collapse of isolated low-mass cloud cores using three-dimensional non-ideal magnetohydrodynamics simulations. In the models explored in this s...
Daisuke Takaishi, Yusuke Tsukamoto et al.
#astro-ph.SR
Prépublication
· 2026 arXiv
EN
We revisit a late-time, model-independent distance-consistency framework that combines strong gravitational lensing systems (SGLS), ultraviolet/X-ray quasars (QSOs), and 38 Sunyaev--Zel'dovich/X-ray galaxy-cluster angular-diameter distances. The analysis focus...
Jian Hu, Yi Liu et al.
#astro-ph.CO
Prépublication
· 2026 arXiv
EN
Gravitational lensing can split the signal of a binary merger into two images, which either overlap and interfere or arrive as separate events. We show that the matched-filter output of every two-image template can be straightforwardly constructed from the sig...
Digvijay Wadekar, Barak Zackay et al.
#gr-qc
Prépublication
· 2026 arXiv
EN
We report a transiting sub-Neptune candidate with a period of 36.34 days around TIC 402710280 (TYC 9041-03677-1), a K dwarf of TESS magnitude 9.24 at 66.8 pc that the Transiting Exoplanet Survey Satellite (TESS) has observed only in full-frame images. Four tra...
Rishit Vora
#astro-ph.EP