Find a story
Search Spins
Search titles, summaries, and missing voices across published articles — press releases, announcements, and media coverage.
7 results for “generative models”
Position: Reasoning is a Learnable Rule-Based Process
A new arXiv position paper argues that autonomous reasoning in AI must be redefined as a learnable, rule-based process grounded in validity and soundness—challenging dominant generative AI paradigms and proposing operational definitions and communication standards to restore construct validity in reasoning evaluation.
Aug 14, 2026
The Fairness Collapse Phenomenon: Bias Amplification in Language Models Trained on Synthetic Data
Researchers identify a new phenomenon—'fairness collapse'—where language models trained recursively on synthetic data amplify social biases faster than they degrade in standard performance metrics, posing a stealth risk to AI equity.
Aug 6, 2026
Computational models of pragmatic reasoning with flexible generation of meaning and expression alternatives
Researchers introduced SAGE, a neuro-symbolic framework that integrates large language models with cognitive modeling to generate and evaluate pragmatic language alternatives, aiming to improve explanatory transparency in computational pragmatics.
Jul 22, 2026
Replicating Belief, Not Bits: Epistemic State Replication for Agentic Systems
Researchers propose Epistemic State Replication (ESR), a new theoretical framework for distributed systems that replaces bitwise state agreement with semantic belief agreement among stochastic, model-driven agents.
Jul 14, 2026
Post-Generation Curation of Synthetic Images via Homogeneous-Heterogeneous Splitting
A new method called Homogeneous-Heterogeneous Splitting improves synthetic image utility by selecting subsets based on fidelity and diversity, without retraining generators, achieving real-data-level performance with up to 40% fewer samples.
Jul 8, 2026
A Filtered Mixture-of-Generators for Fully Synthetic Survival Training
FoGS is a new synthetic data method for survival analysis that improves model performance on scarce clinical data by filtering outputs from multiple generative models using real-data-trained survival scorers, enabling viable real-data substitution in privacy-restricted settings.
Published Jul 2, 2026 · Analyzed Jul 5, 2026
SNAP-FM: Sparse Nonlinear Accelerated Projection for Physics-Constrained Generative Modeling
Researchers propose a new method to accelerate physics-constrained generative modeling.
Published Jul 2, 2026 · Analyzed Jul 5, 2026