Science – Page 24

Global Conversations on Hydrogen: Uncovering Trends Across Languages

22.02.2026 by qfx

Across a decade of analysis (2013-2022), topic modeling reveals evolving themes-identified across multiple languages-that demonstrate the dynamic nature of discourse and the shifting priorities within it.

A new study analyzes a decade of social media data to reveal how discussions about hydrogen energy have evolved regionally and globally.

Data Science by Committee: Automating Exploratory Analysis with Shared Notebooks

21.02.2026 by qfx

A new system harnesses the power of existing data science work to automatically generate more insightful and relevant exploratory data analysis notebooks for new datasets.

The Deduction Gap: Why AI Struggles to Play 20 Questions

21.02.2026 by qfx

The AIDG framework establishes a four-stage pipeline-initialization of models and secrets, iterative interaction between a Seeker and Holder, arbiter-based assessment of information leakage or secure locking, and outcome calculation utilizing Dual-ELO ratings and efficiency weighting-to evaluate and refine security protocols.

New research reveals a surprising imbalance in how artificial intelligence processes information during conversations, consistently excelling at keeping secrets but faltering when asked to uncover them.

Splitting the Inference Load: Privacy and Speed for Large Language Models

21.02.2026 by qfx

A new system efficiently distributes large language model computations between local devices and the cloud to address both privacy concerns and network latency.

Tracking User Intent to Shield AI from Evolving Attacks

21.02.2026 by qfx

The DeepContext architecture processes conversational turns by first embedding them with a fine-tuned BERT model and weighted pooling, then tracking user intent through a GRU-based recurrent network, and finally classifying the resulting conversational trajectory with a hybrid residual connection to assess safety.

New research details a system that monitors conversations with large language models to identify and counter sophisticated, multi-turn adversarial attempts.

Forged in the Machine: Boosting Document Fraud Detection with Synthetic Data

21.02.2026 by qfx

A new approach uses contrastive learning to generate realistic tampered document images, significantly enhancing the performance of forgery detection systems.

Smarter Telecom AI: Grounding Language Models with Dynamic Knowledge

21.02.2026 by qfx

The framework cultivates knowledge from telecom documentation by first using a language model to map entities and their relationships into a knowledge graph, then strategically retrieves pertinent context from this graph to inform a final language model response-a process acknowledging that robust answers don’t arise from isolated data, but from a carefully grown web of interconnected information.

New research demonstrates how integrating constantly updated knowledge graphs with large language models can significantly improve accuracy and trustworthiness in telecom applications.

Seeing the Invisible: AI Pinpoints Faint Gamma-Ray Bursts

21.02.2026 by qfx

The study demonstrates that the ComptonUNet model consistently outperforms alternatives - including standard Unet and models specific to Compton or pinhole imaging - in reconstructing the morphology and peak locations of gamma-ray burst sources across a range of durations, from one to one hundred seconds.

A new deep learning framework dramatically improves the ability to localize fleeting gamma-ray bursts, even with limited and noisy data from Compton cameras.

Sharpening AI Preferences: A New Approach to Reward Modeling

21.02.2026 by qfx

$The system iteratively refines a reward model through adaptive data augmentation, prioritizing samples with minimal preference margin-quantified as [latex]|\Delta_{i}^{t}|[/latex]-to generate synthetic data, thereby focusing learning on the most informative examples and improving model accuracy at each stage [latex]r_{\theta}^{t-1} \rightarrow r_{\theta}^{t}[/latex].$

A novel data augmentation strategy, MARS, improves the reliability and alignment of AI systems by focusing on the most challenging preference comparisons.

The AI Performance Plateau: Why Benchmarks Are Losing Their Edge

21.02.2026 by qfx

As benchmarks age, performance compression among state-of-the-art models increases, leading to a demonstrable rise in mean saturation index-a trend evidenced by older benchmarks exhibiting significantly higher average saturation values, with standard deviation represented within each time bin.

A new study reveals that the rapid gains seen in artificial intelligence are increasingly limited by benchmark saturation, demanding a rethink of how we measure progress.