Weizmann Institute of Science

Systematic differences in response structures reveal that standardized assessments may not accurately reflect LLM capabilities, undermining their validity in AI evaluations.

Traditional curricula may be limiting; learnity graphs could revolutionize how we approach lifelong education by connecting knowledge, skills, and experiences in a dynamic framework.

LeDXA reveals that self-supervised learning can unlock critical health insights from DXA scans, outperforming traditional metrics in predicting disease risk with far fewer resources.

Achieving state-of-the-art video generation with significantly improved diversity and speed, PDD redefines the efficiency of diffusion models.

GLP redefines the landscape of multiagent systems by proving that independent instances can collaborate without central control, embodying true grassroots principles.

LUTRA achieves a 1.25x boost in SNR for undersampled astronomical images, revolutionizing the detection of transient events.

A simple training framework boosts pixel-level tampering detection performance in VLMs by over 26%, showcasing the power of balanced sampling and late-injection strategies.

Optimal tuning is not just beneficial but essential for accurate predictions of electronic fundamental gaps at finite temperatures, fundamentally changing how we approach hybrid functional theory.

Amplifying quantum pseudorandom states to t-copy security without extra assumptions opens new avenues for quantum cryptography.

BabyMind outperforms existing models by leveraging an object-first approach that stabilizes learning in noisy, real-world child-view video data.

Spin-selective coatings can dramatically enhance hydrogen production efficiency by eliminating harmful hydrogen peroxide formation in photoelectrochemical cells.

Dynamic spin processes in chiral molecules could explain why D-handed RNA was favored in the emergence of life, revealing a deeper connection between magnetism and molecular chirality.

Brain-IT-VQA outperforms existing fMRI-based VQA methods, revealing intricate details about how the brain represents visual information.

SSFP-NMR unlocks a new way to dissect complex chemical exchange kinetics, revealing hidden intermediates and their properties even when present in tiny amounts.

F12 explicitly correlated methods don't have to be so computationally expensive: pANO-F12 basis sets offer a route to more compact calculations without sacrificing accuracy.

Decomposing safety proofs into forward, backward, and prophecy steps dramatically simplifies the search for inductive invariants, enabling verification of complex systems like Paxos and Raft.

Even in the noiseless setting of gradient descent, achieving optimal last-iterate convergence without knowing the time horizon is provably impossible, requiring an excess polylogarithmic factor.