arXiv:2602.02543v1 Announce Type: new Abstract: Model editing has emerged as a practical approach for mitigating factual errors and outdated knowledge in large language models (LLMs). Among existing methods, the Locate-and-Edit (L&E) paradigm is the dominant framework: it locates MLP parameters implicated in expressing a target fact, and then performs a localized update to rewrite that fact. However, long sequences of edits often trigger abrupt model collapse in L&E beyond a critical point. We empirically identify a strong correlation […]

Ver mais

arXiv:2602.12534v1 Announce Type: new Abstract: In truncated linear regression, samples $(x,y)$ are shown only when the outcome $y$ falls inside a certain survival set $S^star$ and the goal is to estimate the unknown $d$-dimensional regressor $w^star$. This problem has a long history of study in Statistics and Machine Learning going back to the works of (Galton, 1897; Tobin, 1958) and more recently in, e.g., (Daskalakis et al., 2019; 2021; Lee et al., 2023; 2024). Despite this long history, […]

Ver mais

We aim to develop physics foundation models for science and engineering that provide real-time solutions to Partial Differential Equations (PDEs) which preserve structure and accuracy under adaptation to unseen geometries. To this end, we introduce General-Geometry Neural Whitney Forms (Geo-NeW): a data-driven finite element method. We jointly learn a differential operator and compatible reduced finite element spaces defined on the underlying geometry. The resulting model is solved to generate predictions, while exactly preserving physical conservation laws through Finite […]

Ver mais

LiteLLM Hack: Were You One of the 47,000? Daniel Hnyk used the BigQuery PyPI dataset to determine how many downloads there were of the exploited LiteLLM packages during the 46 minute period they were live on PyPI. The answer was 46,996 across the two compromised release versions (1.82.7 and 1.82.8). They also identified 2,337 packages that depended on LiteLLM – 88% of which did not pin versions in a way that would have avoided the exploited version. Via […]

Ver mais

The accelerating pace of artificial intelligence research and deployment makes both extraordinary opportunity and profound peril increasingly apparent. This paper discusses the innovative proposition that AI can be marsharded, paradoxically, as a proactive guardian of human cognition against the harmful applications of the very technology on which it relies. The heuristic of “brain hacking”—an intentional deployment of AI-driven interventions that systematically augment mental capacities while fortifying neural substrates against adversarial incursions—emerges as a promising trajectory for both theoretical […]

Ver mais

arXiv:2604.15315v1 Announce Type: new Abstract: Consider a two-player game repeated N times. Player 1 can choose between two styles (for interpretability, offensive and defensive), whereas Player 2 uses a single fixed style. Let X N,:= #wins -#losses for Player 1 after N games, and define the match gain as E[sign(X N )], with sign(0) = 0. We assume Player 1 is weaker in the sense that each pure style is losing in expectation. Our objective is to identify […]

Ver mais

Many-Task Learning refers to the setting where a large number of related tasks need to be learned, the exact relationships between tasks are not known. We introduce the Cascaded Transfer Learning, a novel many-task transfer learning paradigm where information (e.g. model parameters) cascades hierarchically through tasks that are learned by individual models of the same class, while respecting given budget constraints. The cascade is organized as a rooted tree that specifies the order in which tasks are learned […]

Ver mais

Manual compliance auditing in cloud environments consumes up to 40% of IT security budgets annually, yet existing approaches verify control presence rather than effectiveness, leaving institutions vulnerable to adversarial evasion. This paper presents an AI-augmented hybrid ML–LLM compliance auditing system evaluated on a national cybersecurity standards framework (143 controls, 200,000 training events). The system combines multi-label XGBoost classification with LLM-based semantic log analysis, grounded in a formal effectiveness model. Key findings: XGBoost achieves 99.88% F1 after 5% domain […]

Ver mais

SQL is still central to modern systems, but much of the work around it is repetitive. Developers spend hours fixing typos, reformatting queries, and rewriting patterns they already know by heart. To be precise, nearly 40% of SQL time goes into these routine tasks every week. That’s a big reason AI tools are spreading so quickly. Today, about 84% of developers already use, or plan to use, AI to handle this kind of work. Instead of wrestling with boilerplate, this helps engineers focus on design, […]

Ver mais