engblogs

MIT Schwarzman College of Computing and MBZUAI have initiated a five-year international collaboration to advance foundational AI research and accelerate its applications in scientific discovery, human thriving, and planetary health.

A recount of Fly.io's CEO falling victim to a sophisticated phishing attack on their legacy Twitter account, highlighting the importance of phishing-resistant authentication and robust security hygiene.

An overview of efforts to detect, disrupt, and prevent malicious uses of AI by threat actors through policy enforcement, public reporting, and collaborative innovation as of October 2025.

Stripe introduces Agentic Commerce Protocol and Shared Payment Tokens to enable secure, programmable, and seamless AI-driven e-commerce experiences while giving businesses control over branding, product data, and fraud protection.

MIT engineers used machine learning-guided simulations and 3D printing to develop a high-strength, temperature-resistant printable aluminum alloy that could revolutionize lightweight aircraft and other advanced applications.

MIT researchers developed a physics-informed machine learning model that accurately predicts plasma behavior during tokamak rampdown, enhancing the safety and reliability of fusion power plants.

Priya Donti leverages AI and machine learning to develop innovative algorithms and synthetic data for optimizing renewable-integrated power grids, aiming to address climate change and social inequity through interdisciplinary collaboration and impactful technology.

Developed an open-source, Apache Phoenix-based DynamoDB-compatible REST service enabling zero-code, multi-cloud database migrations at enterprise scale.

Exploring how Lyft's Rider Experience Score (RES) uses causal inference and machine learning techniques like AIPW to estimate long-term impacts of rider experiences on retention without relying on traditional A/B testing.

An exploration of reward hacking in reinforcement learning models through a custom testbed and fine-tuning experiments, highlighting differences between Qwen and GPT-OSS model families and strategies for monitoring and mitigating exploit behaviors.

Explore five unique Japanese words that encapsulate key cultural values, emotions, and aesthetics, offering deep insight into the Japanese way of thinking and feeling.

AgentKit introduces an integrated suite of tools including Agent Builder, Connector Registry, ChatKit, enhanced Evals, and Reinforcement Fine-Tuning to streamline the design, deployment, evaluation, and optimization of intelligent agent workflows.