Science – Page 16

Smarter Ad Auctions: Boosting Publisher Revenue Through Personalized Pricing

16.01.2026 by qfx

The distribution reveals a clear relationship between average bid and quality score, suggesting that higher quality scores correlate with increased bidding amounts, potentially indicating a willingness to pay a premium for superior performance.

New research details an auction mechanism that leverages user data to significantly increase revenue for digital publishers.

Who’s Drawing What? Unmasking Text-to-Image Model Identities

16.01.2026 by qfx

A system leverages voting-based leaderboards, an adversarial deanonymization pipeline, and model-specific clustering within embedding spaces to establish a comprehensive framework for analyzing and potentially compromising user privacy.

A new study reveals that text-to-image models leave subtle fingerprints in their creations, allowing researchers to reliably identify them even when prompts are unknown.

Unveiling Information’s Hidden Pathways

15.01.2026 by qfx

$The function [latex] g_{\alpha}(x,r) = \ln_{1/\alpha} r(x) [/latex] defines a relationship between a variable <i>r</i> and <i>x</i>, scaled by a parameter α, establishing a logarithmic connection crucial for characterizing the system’s behavior.$

A new framework clarifies the meaning of alpha-mutual information by connecting it to the fundamental concept of information leakage.

The Privacy Instinct: Can AI Truly Understand What We Keep Secret?

15.01.2026 by qfx

The system models individual privacy responses by constructing a memory from user comment history and post context, then uses this to simulate likely reactions to privacy-related events sourced from online technical discussions, with a concern-level judge evaluating alignment against established privacy taxonomies and human-annotated data.

Researchers have developed an AI agent that attempts to model human privacy reasoning, predicting how individuals will react to different privacy-related scenarios.

Turning Vision Models Right-Side Up: Tackling Bias with Simple Rotation

15.01.2026 by qfx

New research reveals a surprisingly effective method for eliminating orientation-induced biases in vision-language and image generation models, leading to fairer and more robust AI.

Seeing Through Deception: A New AI Tackles Image Forgery

15.01.2026 by qfx

Researchers have developed a novel framework that significantly improves the detection of manipulated images, even when those images come from unfamiliar sources.

Seeing Eye-to-Eye with Galaxies: Predicting Interactions with AI

15.01.2026 by qfx

The study demonstrates a comparative analysis of inference speeds-measured in seconds-across four models-H-SENN, AG-XCaps, ResNet-GRU, and an ensemble configuration-highlighting the computational efficiency of each approach.

A new approach leverages the power of artificial intelligence to accurately forecast how galaxies will interact, offering insights into cosmic evolution.

Decoding Danger: Deep Learning Spots Vulnerabilities in Raw Machine Code

15.01.2026 by qfx

Across a spectrum of vulnerability types, both top sequential and graph-based models demonstrate performance, highlighting the potential of these approaches to address diverse security challenges within complex systems.

A new approach leverages deep learning to directly analyze x86-64 machine code, offering a streamlined path to identifying security flaws.

The Deepfake Economy: How AI is Fueling a New Era of Adult Content

15.01.2026 by qfx

$A significant fraction of deepfake requests on Civitai lack the platform’s informational alerts, a deficiency markedly pronounced within non-safe-for-work content, suggesting a systemic gap in intervention despite available tools.$

A new study examines the Civitai platform and its bounty system, revealing a thriving market for AI-generated adult content and the challenges of moderating this rapidly expanding landscape.

Spotting the Sidestep: A New Benchmark for Honest Financial AI

15.01.2026 by qfx

Lower training loss alone is not a reliable indicator of superior test performance, as demonstrated by the contrast between a single-model baseline and Eva-4B, suggesting that the utilization of judge-resolved samples functions as a regularization technique during training.

Researchers have created a new dataset and methodology to better identify when artificial intelligence systems are dodging difficult questions in the realm of financial queries.