How Open Must Language Models be to Enable Reliable Scientific Inference?
arXiv:2603.26539v1 Announce Type: new Abstract: How does the extent to which a model is open or closed impact the scientific inferences that can be drawn from research that involves it? In this paper, we analyze how restrictions on information about model construction and deployment threaten reliab
Development of a European Union Time-Indexed Reference Dataset for Assessing the Performance of Signal Detection Methods in Pharmacovigilance using a Large Language Model
arXiv:2603.26544v1 Announce Type: new Abstract: Background: The identification of optimal signal detection methods is hindered by the lack of reliable reference datasets. Existing datasets do not capture when adverse events (AEs) are officially recognized by regulatory authorities, preventing restr
When Perplexity Lies: Generation-Focused Distillation of Hybrid Sequence Models
arXiv:2603.26556v1 Announce Type: new Abstract: Converting a pretrained Transformer into a more efficient hybrid model through distillation offers a promising approach to reducing inference costs. However, achieving high-quality generation in distilled models requires careful joint design of both t
MemBoost: A Memory-Boosted Framework for Cost-Aware LLM Inference
arXiv:2603.26557v1 Announce Type: new Abstract: Large Language Models (LLMs) deliver strong performance but incur high inference cost in real-world services, especially under workloads with repeated or near-duplicate queries across users and sessions. In this work, we propose MemBoost, a memory-boo
EnTaCs: Analyzing the Relationship Between Sentiment and Language Choice in English-Tamil Code-Switching
arXiv:2603.26587v1 Announce Type: new Abstract: This paper investigates the relationship between utterance sentiment and language choice in English-Tamil code-switched text, using methods from machine learning and statistical modelling. We apply a fine-tuned XLM-RoBERTa model for token-level langua
Weight Tying Biases Token Embeddings Towards the Output Space
arXiv:2603.26663v1 Announce Type: new Abstract: Weight tying, i.e. sharing parameters between input and output embedding matrices, is common practice in language model design, yet its impact on the learned embedding space remains poorly understood. In this paper, we show that tied embedding matrice
Do Neurons Dream of Primitive Operators? Wake-Sleep Compression Rediscovers Schank's Event Semantics
arXiv:2603.25975v1 Announce Type: cross Abstract: We show that they do. Schank's conceptual dependency theory proposed that all events decompose into primitive operations -- ATRANS, PTRANS, MTRANS, and others -- hand-coded from linguistic intuition. Can the same primitives be discovered automatical
Policy-Guided World Model Planning for Language-Conditioned Visual Navigation
arXiv:2603.25981v1 Announce Type: cross Abstract: Navigating to a visually specified goal given natural language instructions remains a fundamental challenge in embodied AI. Existing approaches either rely on reactive policies that struggle with long-horizon planning, or employ world models that su
H-Node Attack and Defense in Large Language Models
arXiv:2603.26045v1 Announce Type: cross Abstract: We present H-Node Adversarial Noise Cancellation (H-Node ANC), a mechanistic framework that identifies, exploits, and defends hallucination representations in transformer-based large language models (LLMs) at the level of individual hidden-state dim
Semi-Automated Knowledge Engineering and Process Mapping for Total Airport Management
arXiv:2603.26076v1 Announce Type: cross Abstract: Documentation of airport operations is inherently complex due to extensive technical terminology, rigorous regulations, proprietary regional information, and fragmented communication across multiple stakeholders. The resulting data silos and semanti
Selective Deficits in LLM Mental Self-Modeling in a Behavior-Based Test of Theory of Mind
arXiv:2603.26089v1 Announce Type: cross Abstract: The ability to represent oneself and others as agents with knowledge, intentions, and belief states that guide their behavior - Theory of Mind - is a human universal that enables us to navigate - and manipulate - the social world. It is supported by
Finding Distributed Object-Centric Properties in Self-Supervised Transformers
arXiv:2603.26127v1 Announce Type: cross Abstract: Self-supervised Vision Transformers (ViTs) like DINO show an emergent ability to discover objects, typically observed in [CLS] token attention maps of the final layer. However, these maps often contain spurious activations resulting in poor localiza
DataFlex: A Unified Framework for Data-Centric Dynamic Training of Large Language Models
arXiv:2603.26164v1 Announce Type: cross Abstract: Data-centric training has emerged as a promising direction for improving large language models (LLMs) by optimizing not only model parameters but also the selection, composition, and weighting of training data during optimization. However, existing
Research
Watch SpaceX launch 119 payloads to orbit from California early on March 30
SpaceX will launch 119 payloads to orbit from California early Monday morning (March 30), and you can watch the action live.
Research
'We are ready': NASA still on track to launch Artemis 2 astronauts to the moon April 1
NASA has a high degree of confidence that its historic Artemis 2 crewed moon mission will launch on time.
Research
Saturn's chaotic atmosphere revealed in most comprehensive view yet by James Webb and Hubble telescopes
Viewing Saturn in complementary wavelengths, the James Webb and Hubble space telescopes reveal more about what makes up the layers of ringed planet's atmosphere.
Research
'Warhammer 40,000: Dawn of War 4': Release date, plot, factions, & everything else we know about the upcoming 40K strategy game
In the grim darkness of the future, there's only real-time strategy as the venerable Dawn of War series returns for a fourth entry.
Research
AI systems are enabling mass surveillance in the US, and there is no national law that 'meaningfully limits' the use of this data
A technology policy researcher explores the ethics of implementing AI in current camera surveillance systems.
Research
Endometriosis messes with the immune system and causes 'ripple effects across the body'
Two researchers explain how endometriosis drives a whole body immune response due to inflammation, urging the medical community to see it as a whole-body issue.
Research
Kristen Stewart will play Sally Ride, the 1st American woman in space, as Amazon finally greenlights mini-series 'The Challenger'
In herTV series debut, 'The Twilight Saga' star will portray the first American woman in space.