Can AI Fix Your Rust Code?
Researchers introduce a new benchmark and agentic framework to dramatically improve automated bug fixing in Rust programs.
Science
Decoding the Stability of AI Minds
![The system’s stability hinges on a margin-to-degeneracy geometry, where each token’s proximity to an instability boundary-analogous to a support vector in [latex]SVM[/latex] classification-dictates the sequence’s overall resilience, with the token closest to this boundary exerting disproportionate influence on the system’s susceptibility to perturbation.](https://arxiv.org/html/2602.22271v1/2602.22271v1/figures/figure_margin_schematic.png)
New research reveals a geometric principle underlying the robustness of large language models, offering a deeper understanding of how they learn and generalize.
Outsmarting the Bot: How to Secure Language Models Against Malicious Prompts
New research reveals the strengths and weaknesses of current defenses against prompt injection and jailbreak attacks targeting large language models.
Unlocking the Secrets of Material Conductivity
![The half-filled square-lattice Hubbard model exhibits a complex crossover from a weak-coupling Fermi liquid to a strong-coupling Mott insulator, punctuated by an intervening bad metal phase-identified through the local maximum of thermal entropy-and further characterized by the alignment of inflection points in double occupancy, the peak position of the AFM structure factor, the disappearance of quasiparticle coherence, minima in thermal entropy, fidelity susceptibility peaks, and crossings in imaginary self-energy components, all converging near a low-temperature transition from Slater to Mott insulator at [latex]U^{\*}/t \sim eq 4.25 [/latex].](https://arxiv.org/html/2602.22705v1/2602.22705v1/x1.png)
New simulations reveal a detailed understanding of how materials transition between conducting and insulating states, paving the way for better material design.
Seeing Clearly Through the Noise: A New Approach to Multi-View Clustering
Researchers have developed a framework that improves data clustering accuracy when dealing with noisy and incomplete information from multiple sources.
Untangling Quantum Fields: A New Route to Confinement
![The qubit-regularized model, governed by the Hamiltonian [latex] \tilde{1} [/latex], suggests that a substantial continuum quantum field theory can arise from a carefully chosen perturbation at its quantum critical point, hinting at emergent behavior from a fundamentally discrete system.](https://arxiv.org/html/2602.22515v1/2602.22515v1/x4.png)
Researchers are leveraging qubit-based simulations to explore the fundamental phases of matter described by lattice gauge theories and potentially unlock access to complex quantum field theories.
Beyond Vectors: Modeling Language with the Laws of Physics
A new approach to sequence modeling leverages principles from quantum mechanics to represent language, offering potential advantages in both expressive power and data efficiency.
Squeezing More from Less: A New Approach to Compact Language Models
Researchers have developed a novel training method that dramatically reduces the size of language models without sacrificing accuracy.
Beating the Heat: Phosphors for Brighter, More Efficient Lighting
This review explores the latest advancements in metal halide phosphors designed to maintain luminescence at high temperatures, paving the way for next-generation displays and lighting technologies.
Keeping Calculations on Track: Fault Tolerance for Distributed Operations
This review explores techniques for ensuring reliable results in distributed computing environments when processes inevitably fail during collective operations.