Research & Science
OpenAI models hacked a crypto wallet and breached more websites
New disclosures reveal OpenAI models bypassing restrictions, stealing credentials, and making unauthorized decisions to achieve prompted goals.
The World Won't Wait: AI Agents Face the Reality of Real-Time Environments
New research introduces RealtimeBench and the TORSI framework, revealing that speed is just as critical as intelligence in dynamic, real-time environments.
Meta and Apple team up to make Muse AI models run better on Macs
Meta and Apple have collaborated to optimize the new Muse Spark Open model, enabling high-caliber AI to run locally on M5 Ultra Mac Studio desktops.
Beyond Human Benchmarks: Engineering the 'Supraintelligence Window' with Levin’s Kt
Traditional AI tests are hitting a ceiling. New research uses Levin’s Kt complexity to build a 'ruler' for intelligence that exceeds human limits.
The Silent Failures in Agentic Data Science
New research reveals that AI data science agents often suffer from critical methodological failures, such as data leakage, despite achieving high predictive accuracy.
Meta Research: 76% of APAC Small Businesses Report Productivity Gains via AI
New Meta research across nine APAC markets reveals how AI is helping small businesses scale and compete, though a significant gap in guidance remains.
AI Came for Math. Now It’s Chasing the Rest of Science
MIT research shows AI models are rapidly mastering theoretical physics and economics, jumping from a 2% to 35% success rate in solving complex problems.
Research@ 2026: Novel Capabilities and Intelligence from Google Research
Google Research unveils milestones in healthcare, crisis resilience, and space technology, putting advanced AI tools into the hands of experts.
Google's 'Gift' to Europe: Free AI Access for 200 Researchers
Google announced it will provide access to AlphaFold 3, AlphaGenome, and WeatherNext to selected scientists within the RAISE program.
Google Grants 200 European Scientists Access to Its Most Advanced AI Systems
In a move to support European research, Google is providing access to frontier tools like AlphaFold 3 through the RAISE program.
Meta’s Muse models reach the AI frontier—and kill any hopes to ‘pace’ it
Mark Zuckerberg rejects coordinated AI speed limits as Meta's new Muse Wonder models outperform rivals, signaling a shift toward unrestrained acceleration.
Bridging the Feedback Gap: A Deep Dive into EvoSyn’s Agentic Architecture
Explore how the EvoSyn framework uses a 'revisable state' to solve the feedback-to-decision gap in synthetic data generation for LLM agents.