Science – Page 6

Beyond Benchmarks: Automating Foundation Model Evaluation

06.12.2025 by pricpr

The TASKEVAL system introduces a framework for task evaluation built upon a task interaction protocol and a task-agnostic meta-model, culminating in an Eval Synthesiser designed to assess performance across diverse challenges-a configuration reflecting an acceptance of inevitable systemic evolution rather than static perfection.

A new framework, TaskEval, offers a scalable solution to the challenge of consistently and accurately assessing the performance of large AI models across diverse tasks.

Atomic Focus: Boosting RF Reception with 3D-Printed Optics

06.12.2025 by pricpr

Electromagnetic interference was demonstrably enhanced through the implementation of a GRIN Luneburg-type metamaterial lens, as evidenced by increased signal strength within the electromagnetically induced transparency (EIT) window at both $2.2$ GHz and $3.6$ GHz, indicating a pathway toward more sensitive Rydberg atom receivers.

Researchers have significantly enhanced the sensitivity of quantum-based radio frequency receivers by focusing signals onto a vapor cell using a custom-designed metamaterial lens.

Mapping the Milky Way’s Dark Side

06.12.2025 by pricpr

New simulations and machine learning techniques are refining our understanding of dark matter’s distribution within our galaxy, paving the way for more sensitive direct detection experiments.

Beyond the Limits of Quantum Coherence

06.12.2025 by pricpr

New research explores the surprising potential for amplifying a fragile quantum property, and defines the boundaries of its enhancement under realistic constraints.

Beyond Equilibrium: Universal Signatures of Quantum Phase Transitions

06.12.2025 by pricpr

$The study of the dissipative Dicke model reveals that near the critical point, the thermodynamic metric $ζ$ scales according to $ζ \sim |g - g_{c}|^{x}$, with a numerically determined exponent of $x = -2.9953 \pm 0.0009$ - a value remarkably consistent with the theoretical prediction of $x = -(\alpha + \gamma)$, given parameters $\alpha = 2$ and $\gamma = 1$.$

New research reveals a unifying principle governing how quantum systems transition between phases, even when driven far from equilibrium.

Beyond Hermiticity: A Dissipative Spin Liquid Model Unveiled

06.12.2025 by pricpr

$The study demonstrates that the Liouvillian spectrum-analyzed for systems with linear sizes ranging from 4 to 64-reveals a phase transition point, $\gamma_{PT}$, which scales inversely with system size ($1/L$), while a critical value, $\gamma^*$, remains approximately constant at $3J/4$, and the fraction of broken $\mathcal{PT}$ symmetry smoothly transitions from 0 to 1 as the parameter $\gamma$ crosses these thresholds.$

Researchers have discovered an exactly solvable model demonstrating how non-Hermitian dynamics can give rise to exotic quantum phenomena in dissipative spin liquids.

When Perturbation Fails: Taming Light in Complex Cavities

06.12.2025 by pricpr

$A ring cavity, incorporating a dispersing medium, modulates the propagation of a pulsed light field $E_i(t,z)$ pumped through an input-output mirror with reflection coefficient $\sqrt{\mathcal{R}}$ and transmission coefficient $\sqrt{\mathcal{T}}$, effectively creating a round-trip time $T_{R}$ that governs the cavity’s dispersive response.$

A new analysis reveals the limits of simplified modeling techniques used to predict how light pulses behave within dispersive optical cavities.

Smarter Inference: Guiding Language Models with Real-Time Rewards

06.12.2025 by pricpr

A new approach selectively builds upon promising generations during inference, boosting performance beyond traditional methods.

Bridging Microwaves and Light: A New Path to Quantum Networks

06.12.2025 by pricpr

Researchers demonstrate a promising pump-free method for converting microwave signals into optical photons, a critical step towards building long-distance quantum communication systems.

Untangling Qubit Control: A Bayesian Approach

06.12.2025 by pricpr

$Stepwise two-parameter estimation, utilizing prior distributions with a width of $5^{\circ}$, demonstrates that the accuracy of shift estimation-indicated by a standard deviation of $\Delta\gamma$-approaches the classical Van Trees limit as a function of angle $\theta$, while axis estimation accuracy, similarly assessed with $\Delta\theta$, closely aligns with the theoretical prediction and remains consistent with the diagonal, signifying effective parameter recovery.$

New research explores a practical method for precisely calibrating multiple qubit rotations using Bayesian inference.