Prépublication
· 2026 arXiv
EN
We present a model of slow-roll inflation in metric-affine gravity, where the connection is independent of the metric. We require that the Einstein-Hilbert term is the only operator that breaks three out of four shift-symmetries of the connection, and impose t...
Christian Dioguardi
#gr-qc
Prépublication
· 2026 arXiv
EN
Pixel-space diffusion models have recently re-emerged as a compelling alternative to latent diffusion, avoiding the lossy latent bottleneck and separately trained autoencoder. JiT shows that a vanilla diffusion transformer can generate directly in pixel space...
Soumik Mukhopadhyay, Long Mai et al.
#cs.CV
Prépublication
· 2026 arXiv
EN
Fine-tuning large language models (LLMs) can induce behaviors beyond those explicitly targeted. For example, training a model to respond more warmly can make it more sycophantic, and training it to write insecure code can cause broad misalignment on prompts un...
Dongyoon Hahm, Woogyeol Jin et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Reproducible and analytically robust research findings are essential to scientific progress, but evaluating these properties typically requires costly manual effort. We investigate whether LLM agents can help automate reproduction and robustness analysis, usin...
Shicheng Liu, Krista Opsahl-Ong et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
Scientific discovery requires developing worthwhile hypotheses into practical investigations, yet learning from this evolving research process remains challenging. We introduce an operational definition of valuable and testable research hypotheses: evidence-gr...
Jianyuan Zhong, Chenwei Wu et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
A reasoning trace can accompany a correct answer without providing effective supervision for a particular student. We study how to construct traces that improve learning beyond answer-only supervision by considering both their intermediate content and their ac...
Shaoang Li, Yanhang Shi et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Self-driving laboratories (SDLs) accelerate scientific discovery by automating experimental execution and optimization. AI agents can increasingly propose new experiments, but existing SDLs can only run the procedures they were built for. Adapting them to a ne...
Jiaru Bai, Yuchi Zhao et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
Speculative decoding typically characterizes speedup by draft tokens accepted per verification round. Lossless verification preserves the target model's output distribution and thus expected completion length, making this metric a meaningful efficiency proxy....
Chiatzen Wang, Bill Xu et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
While deriving text embeddings directly from frozen LLMs has emerged as a promising direction, existing methods focus almost exclusively on English, leaving multilingual settings largely unexplored. In this work, we present the first systematic study of multil...
Ailiang Lin, Zhuoyun Li et al.
#cs.CL
Prépublication
· 2026 arXiv
EN
Sparse autoencoders (SAEs) explain foundation model representations by decomposing activations into sparse combinations of learned dictionary atoms. However, the learned dictionaries can vary across training runs, limiting the reliability and reproducibility o...
Jinho Choi, Hyesu Lim et al.
#cs.AI
Prépublication
· 2026 arXiv
EN
Transformer prefill reuses weights across tokens and is dominated by matrix multiplication and activation data movement; single-token decode has limited reuse and is dominated by weight and KV-cache reads. We present FUTRANSFORMER, a CUDA implementation of a L...
Wei Wang
#cs.AI
Prépublication
· 2026 arXiv
EN
Computational drug discovery requires agents to compose scientific operations and adapt to changing evidence and constraints. Here we introduce DrugHarness, an evolvable agent harness coupling composable scientific execution with conditional skill induction. I...
Yuxing Chen, Zhisheng Zhang et al.
#cs.AI