Unlocking Harm: How Sentence Pairing Exposes Language Model Weaknesses
![SPELL streamlines large language model serving through speculative decoding, pre-drafting potential continuations [latex] \hat{y}_{t} [/latex] with a small language model before verifying them with a larger, more accurate one, thereby reducing latency and improving throughput even under high load.](https://arxiv.org/html/2512.21236v1/figures/overview.png)
New research reveals a surprisingly effective method for prompting large language models to generate malicious code, even those considered highly secure.
![SPELL streamlines large language model serving through speculative decoding, pre-drafting potential continuations [latex] \hat{y}_{t} [/latex] with a small language model before verifying them with a larger, more accurate one, thereby reducing latency and improving throughput even under high load.](https://arxiv.org/html/2512.21236v1/figures/overview.png)
New research reveals a surprisingly effective method for prompting large language models to generate malicious code, even those considered highly secure.

A novel mathematical approach establishes a direct link between the structure of information networks and the principles of intuitionistic logic.

Researchers have developed a novel neural speech codec that significantly reduces bandwidth requirements without sacrificing audio quality.

Researchers have developed a new framework, ESCHER, to efficiently represent and analyze evolving hypergraphs, unlocking significant speedups for key graph computations.
A new study explores whether large language models can seamlessly continue mathematical problem-solving started by another AI, revealing significant performance differences based on model architecture.
A new method breaks down the complex problem of assessing network resilience to link uncertainties into smaller, more manageable components.

A new hierarchical modeling approach dramatically improves the speed and memory efficiency of processing lengthy text sequences.

Researchers have demonstrated a practical method to evade radio frequency-based drone detection systems by subtly manipulating the drone’s emitted signals.
![A linear optical network, built upon the [latex]D_{4}^{2}[/latex] Dicke digraph, undergoes stepwise decomposition, with the relative positions of its 44-partite and 33-partite ports intentionally reversed to enhance clarity of the underlying network structure.](https://arxiv.org/html/2512.20881v1/Figures_main/success_prob.jpg)
Researchers have devised a new method for generating complex entangled states, known as Dicke states, using a streamlined optical approach.
Researchers have developed a method to implement encrypted control for complex nonlinear systems by approximating dynamics with autoregressive models.