arXiv:2510.00133v1 Announce Type: new Abstract: Foundational models based on the transformer architecture are currently the state-of-the-art in general language modeling, as well as in scientific areas such as material science and climate. However, training and deploying these models is computationally…
arXiv:2509.06608v3 Announce Type: replace Abstract: The mechanisms by which reasoning training reshapes LLMs’ internal computations remain unclear. We study lightweight steering vectors inserted into the base model’s residual stream and trained with a reinforcement-learning objective. These vectors match full fine-tuning…
arXiv:2510.00136v1 Announce Type: new Abstract: We are born with the ability to learn concepts by comparing diverse observations. This helps us to understand the new world in a compositional manner and facilitates extrapolation, as objects naturally consist of multiple concepts.…
arXiv:2501.14615v3 Announce Type: replace-cross Abstract: Calcium imaging allows for the parallel measurement of large neuronal populations in a spatially resolved and minimally invasive manner, and has become a gold-standard for neuronal functionality. While deep generative models have been successfully applied…
arXiv:2510.00144v1 Announce Type: new Abstract: The ability of reinforcement learning algorithms to learn effective policies is determined by the rewards available during training. However, for practical problems, obtaining large quantities of reward labels is often infeasible due to computational or…
arXiv:2509.25339v2 Announce Type: replace-cross Abstract: Is basic visual understanding really solved in state-of-the-art VLMs? We present VisualOverload, a slightly different visual question answering (VQA) benchmark comprising 2,720 question-answer pairs, with privately held ground-truth responses. Unlike prior VQA datasets that typically…
arXiv:2510.00163v1 Announce Type: new Abstract: The wide adoption of AI decision-making systems in critical domains such as criminal justice, loan approval, and hiring processes has heightened concerns about algorithmic fairness. As we often only have access to the output of…
arXiv:2510.00565v1 Announce Type: cross Abstract: Diffusion language models (DLMs) generate tokens in parallel through iterative denoising, which can reduce latency and enable bidirectional conditioning. However, the safety risks posed by jailbreak attacks that exploit this inference mechanism are not well…
arXiv:2510.00184v1 Announce Type: new Abstract: Language models are increasingly capable, yet still fail at a seemingly simple task of multi-digit multiplication. In this work, we study why, by reverse-engineering a model that successfully learns multiplication via emph{implicit chain-of-thought}, and report…
arXiv:2510.00823v1 Announce Type: cross Abstract: The recently proposed Broximal Point Method (BPM) [Gruntkowska et al., 2025] offers an idealized optimization framework based on iteratively minimizing the objective function over norm balls centered at the current iterate. It enjoys striking global…
