Research - News Articles

Harvard Study Finds AI Outperforms Doctors in Emergency Triage and Treatment Planning, But Experts Urge Caution

Apr 30, 2026

the Guardian

A groundbreaking Harvard study published in Science finds AI outperforms doctors in emergency triage with 67% diagnostic accuracy versus 50-55% for physicians, and dominates long-term treatment planning at 89% accuracy compared to just 34% for human doctors, though researchers caution that unresolved safety and accountability concerns mean AI is not yet ready to replace physicians in routine clinical settings.

Healthcare Foundation Models Research

Harvard Study Finds OpenAI's o1 AI Matches or Beats ER Doctors in Triage and Diagnosis

Apr 30, 2026

Harvard Magazine

OpenAI's o1 AI model matches or outperforms ER doctors in triage, diagnosis, and clinical case management across 76 real emergency cases, according to a Harvard-led study published in Science, with researchers urging urgent clinical trials to explore how AI can partner with physicians to reduce errors and improve patient outcomes.

Healthcare Foundation Models Research

MIT, WPI, and Google Introduce WRING Technique to Reduce AI Bias Without Creating New Ones

Apr 30, 2026

MIT News | Massachusetts Institute of Technology

MIT, WPI, and Google researchers unveil WRING, a groundbreaking debiasing technique that eliminates bias in AI vision-language models by rotating bias-responsible coordinates in high-dimensional space, avoiding the 'Whac-A-Mole dilemma' of creating new biases while fixing old ones.

Research Ethics Computer Vision

Google DeepMind Launches ProEval, Cutting AI Evaluation Costs by Up to 100x With Open-Source Bayesian Tool

Apr 30, 2026

GitHub

Google DeepMind launches ProEval, a free open-source tool that slashes generative AI evaluation costs by up to 100x using Bayesian Quadrature techniques, achieving ±1% accuracy with a fraction of typical samples while proactively identifying model failure patterns across major benchmarks.

Research Open Source Foundation Models

New AI Framework LaDiR Outperforms Existing Reasoning Models Across Math and Planning Benchmarks

Apr 30, 2026

Apple Machine Learning Research

A groundbreaking new AI reasoning framework called LaDiR is outperforming all existing models across math and planning benchmarks by combining latent diffusion models with continuous representations to generate more accurate, diverse, and interpretable reasoning trajectories.

Research Foundation Models

MIT and IBM Launch New Computing Research Lab, Expanding Collaboration Into Quantum and Hybrid Systems

Apr 30, 2026

MIT News | Massachusetts Institute of Technology

MIT and IBM are launching the MIT-IBM Computing Research Lab, expanding their seven-year AI collaboration into quantum and hybrid computing systems to tackle real-world challenges like drug discovery, weather prediction, and financial forecasting.

Quantum Computing Research Science

Amateur Uses Single ChatGPT Prompt to Crack 60-Year-Old Math Conjecture That Stumped World-Class Mathematicians

Apr 27, 2026

Scientific American

A 23-year-old amateur with no advanced math training cracks a 60-year-old mathematical conjecture using a single ChatGPT prompt, producing a novel proof that stumped world-class mathematicians for decades and is now being refined by leading experts like Terence Tao.

Science Foundation Models Research

NVIDIA Launches Open-Source Project Lyra, Enabling AI-Powered 3D World Generation From a Single Image

Apr 27, 2026

GitHub

NVIDIA's Spatial Intelligence Lab launches Project Lyra, a groundbreaking open-source series of generative 3D world models capable of creating explorable, 3D-consistent environments from a single image, with Lyra-2.0 now live and the GitHub repository already gaining nearly 1,800 stars.

Image & Video Open Source Hardware Research

New FlashDrive Framework Cuts Autonomous Driving AI Response Time by 4.5× to Under 200ms

Apr 21, 2026

Z Lab

A breakthrough AI framework called FlashDrive slashes autonomous driving response times from 716ms to just 159ms — a 4.5× speedup — making real-time deployment of Vision-Language-Action models viable across multiple NVIDIA platforms with negligible accuracy loss.

Autonomous Vehicles Research Hardware

New Benchmark Exposes Hidden 'Flinch' Effect in AI Models That Suppresses Words at Probability Level, Defying Uncensoring Fixes

Apr 21, 2026

Morgin.ai

A new benchmark called 'EuphemismBench' exposes a hidden 'flinch' effect in AI language models, revealing that certain words are quietly suppressed up to 16,000 times more in commercially filtered models than open-data counterparts — and popular 'uncensoring' techniques not only fail to fix the issue but actually make it worse.

Foundation Models AI Safety Research