engblogs

Exploring how AI technologies enable Northern Ireland teachers to reclaim valuable time and improve efficiency.

A concise, one-line exploration of prompt injections in AI, outlining attack surfaces and layered defenses, including sandboxing, red-teaming, and the Instruction Hierarchy to protect user data and intent.

Explores the emerging security threat of prompt injections in conversational AI, detailing attack examples, defense strategies, and ongoing research to safeguard user data and maintain AI reliability.

Terminal-Bench 2.0 advances AI agent evaluation with 89 rigorously verified, challenging command-line tasks and introduces Harbor for scalable containerized testing environments.

MIT Energy Initiative has launched the Data Center Power Forum to unite researchers and industry experts in developing sustainable, efficient energy solutions to meet the rapidly growing data center power demands fueled by AI expansion.

The Inworld TTS 1 Max model, powered by Modular's platform, achieves top ranking on the Artificial Analysis speech leaderboard due to its advanced multilingual capabilities, efficient transformer-based architecture, and accelerated API performance delivering low latency and cost-effective text-to-speech synthesis.

The Teen Safety Blueprint outlines a responsible AI framework designed to protect and empower teens through age-appropriate features, safeguards, and ongoing evaluation.

AI is advancing rapidly with the potential to transform knowledge discovery, augment human capabilities, and impact society profoundly, while necessitating robust safety measures, shared regulations, and coordinated governance to ensure beneficial outcomes.

BBVA is transforming its global operations by scaling AI adoption through secure governance, extensive training, and widespread deployment of Custom GPTs, driving significant efficiency and productivity gains across the organization.

MIT CSAIL researchers introduce a modular software model using 'concepts' and 'synchronizations' to enhance code legibility, reliability, and AI-assisted development.

MIT-IBM Watson AI Lab students are advancing AI reliability, efficiency, and multimodal reasoning through innovative techniques in trustworthy model probing, knowledge-grounded reinforcement learning, transformer optimization, and synthetic visual dataset generation.

Spotify enhanced its Fleet Management system with AI-driven background coding agents, automating over 1,500 complex pull requests to significantly boost developer productivity and streamline large-scale codebase maintenance.