AI Research

Leaders are asking the wrong question about AI by seeking forecasts, when the technology's trajectory is unknowable. Resilience, which aims to restore a prior state, is a losing strategy; instead, organizations must adop…

Google Research introduced TabFM, a foundation model for tabular data that performs zero-shot classification and regression without dataset-specific training. The model uses a hybrid-attention architecture combining TabP…

The Rubin Observatory has begun a 10-year mission to create a 'cosmic movie' of the sky, capturing repeated images to track time-domain changes. This long-term survey will generate multi-epoch datasets valuable for trans…

DeepReinforce released Ornith 1.0, an open-source AI model for agentic coding that writes its own scaffolding during training. Unlike traditional reinforcement learning setups, the model proposes improved scaffolds befor…

A new analysis explores how weight quantization affects model performance and sparse autoencoder (SAE) reconstruction of the residual stream, using fraction of variance unexplained (FVU) as a key metric. The study, inspi…

Researchers introduced FlexViT, a reconfigurable FPGA accelerator that achieves up to 2.74x speedup for Vision Transformer layers on edge devices. The accelerator uses a hardware-software co-design approach with a dual-m…

Researchers have developed a new method using approximate leave-one-out (ALO) estimators to accelerate conformal prediction, maintaining accuracy while drastically reducing computational time. The approach promises to en…

Researchers introduced AutoBackSwap, a method that reduces AI image classifiers' reliance on spurious background correlations by automatically swapping backgrounds in training data. The approach improves model robustness…

Researchers developed the Physics-aware Neural Operator Transformer (PNOT), an AI model that enables real-time temperature control in fusion devices by reconstructing divertor temperature fields faster and more accuratel…

A new study shows that high-fidelity wave-based acoustic simulations reduce word error rates in AI speech enhancement by up to 38% compared to lower-fidelity methods. The research, published in Japanese, trained the Spat…

Researchers have identified linguistic bias as a critical vulnerability in voice biometrics spoofing detectors, which perform poorly on diverse real-world data. A new framework using teacher-student adversarial learning …

Researchers have developed a domain-decomposed neural network framework that solves partial differential equations on unbounded domains with improved accuracy and flexibility. The method assigns separate randomized subne…

Researchers have developed a new approach using the Artifact-Transform Workflow Language (ATWL) and an AI assistant that reduces visual-analytics testing from months to hours. The method, which involves relaxing the Pare…

Researchers have developed a new framework for detecting backdoor attacks in large language models, addressing the challenge of discrete input spaces. The framework introduces Class Subspace Orthogonalization (CSO) to en…

Researchers have developed a compact fingerprint model for AI skill identity that uses locality-sensitive hashing to preserve skill relationships, achieving 0.974 AUC on pairwise comparisons with 77 times fewer bits than…

Researchers have developed a failure-driven self-improvement loop for AI training that leverages failed trajectories to boost performance, challenging the traditional focus on successes. The approach improved the OpenCUA…

A new unified AI model integrates textual reasoning and graph processing into a single masked diffusion language model, outperforming graph neural networks, graph transformers, and LLM-based baselines by up to 3.9 points…

Loop engineering is emerging as a critical discipline for building production-ready AI agents, comparable to data structures in traditional software development. The article outlines a roadmap for mastering loop engineer…

Anthropic's new Claude Sonnet 5 outperformed its flagship Opus 4.8 in tests conducted by a user, challenging expectations for midsize models. The results highlight the rapid advancement of AI model capabilities.

A 2025 randomized controlled trial by METR found that experienced developers using AI tools were about 19% slower, despite expecting a 24% speedup. In contrast, a team at fortiss built the Punctilious Platform with a str…